Add show improvement features: crossfade, emotions, returning callers, transcripts, screening
- Music crossfade: smooth 3-second blend between tracks instead of hard stop/start - Emotional detection: analyze host mood from recent messages so callers adapt tone - AI caller summaries: generate call summaries with timestamps for show history - Returning callers: persist regular callers across sessions with call history - Session export: generate transcripts with speaker labels and chapter markers - Caller screening: AI pre-screens phone callers to get name and topic while queued Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
449
backend/main.py
449
backend/main.py
@@ -25,6 +25,7 @@ from .services.llm import llm_service
|
||||
from .services.tts import generate_speech
|
||||
from .services.audio import audio_service
|
||||
from .services.news import news_service, extract_keywords, STOP_WORDS
|
||||
from .services.regulars import regular_caller_service
|
||||
|
||||
app = FastAPI(title="AI Radio Show")
|
||||
|
||||
@@ -115,7 +116,8 @@ CALLER_BASES = {
|
||||
|
||||
|
||||
def _randomize_callers():
|
||||
"""Assign random names and voices to callers, unique per gender."""
|
||||
"""Assign random names and voices to callers, unique per gender.
|
||||
Overrides 2-3 slots with returning regulars when available."""
|
||||
num_m = sum(1 for c in CALLER_BASES.values() if c["gender"] == "male")
|
||||
num_f = sum(1 for c in CALLER_BASES.values() if c["gender"] == "female")
|
||||
males = random.sample(MALE_NAMES, num_m)
|
||||
@@ -125,6 +127,8 @@ def _randomize_callers():
|
||||
f_voices = random.sample(female_pool, min(num_f, len(female_pool)))
|
||||
mi, fi = 0, 0
|
||||
for base in CALLER_BASES.values():
|
||||
base["returning"] = False
|
||||
base["regular_id"] = None
|
||||
if base["gender"] == "male":
|
||||
base["name"] = males[mi]
|
||||
base["voice"] = m_voices[mi]
|
||||
@@ -134,6 +138,32 @@ def _randomize_callers():
|
||||
base["voice"] = f_voices[fi]
|
||||
fi += 1
|
||||
|
||||
# Override 2-3 random slots with returning callers
|
||||
try:
|
||||
returning = regular_caller_service.get_returning_callers(random.randint(2, 3))
|
||||
if returning:
|
||||
keys_by_gender = {"male": [], "female": []}
|
||||
for k, v in CALLER_BASES.items():
|
||||
keys_by_gender[v["gender"]].append(k)
|
||||
|
||||
for regular in returning:
|
||||
gender = regular["gender"]
|
||||
candidates = keys_by_gender.get(gender, [])
|
||||
if not candidates:
|
||||
continue
|
||||
key = random.choice(candidates)
|
||||
candidates.remove(key)
|
||||
base = CALLER_BASES[key]
|
||||
base["name"] = regular["name"]
|
||||
base["returning"] = True
|
||||
base["regular_id"] = regular["id"]
|
||||
# Keep the randomly assigned voice — regulars sound different each time
|
||||
if returning:
|
||||
names = [r["name"] for r in returning]
|
||||
print(f"[Regulars] Injected returning callers: {', '.join(names)}")
|
||||
except Exception as e:
|
||||
print(f"[Regulars] Failed to inject returning callers: {e}")
|
||||
|
||||
_randomize_callers() # Initial assignment
|
||||
|
||||
# Background components for dynamic generation
|
||||
@@ -1239,10 +1269,68 @@ def pick_location() -> str:
|
||||
return random.choice(LOCATIONS_OUT_OF_STATE)
|
||||
|
||||
|
||||
def _generate_returning_caller_background(base: dict) -> str:
|
||||
"""Generate background for a returning regular caller."""
|
||||
regular_id = base.get("regular_id")
|
||||
regulars = regular_caller_service.get_regulars()
|
||||
regular = next((r for r in regulars if r["id"] == regular_id), None)
|
||||
if not regular:
|
||||
return generate_caller_background(base)
|
||||
|
||||
gender = regular["gender"]
|
||||
age = regular["age"]
|
||||
job = regular["job"]
|
||||
location = regular["location"]
|
||||
traits = regular.get("personality_traits", [])
|
||||
|
||||
# Build previous calls section
|
||||
prev_calls = regular.get("call_history", [])
|
||||
prev_section = ""
|
||||
if prev_calls:
|
||||
lines = [f"- {c['summary']}" for c in prev_calls[-3:]]
|
||||
prev_section = "\nPREVIOUS CALLS:\n" + "\n".join(lines)
|
||||
prev_section += "\nYou're calling back with an update — something has changed since last time. Reference your previous call(s) naturally."
|
||||
|
||||
# Reuse standard personality layers
|
||||
interest1, interest2 = random.sample(INTERESTS, 2)
|
||||
quirk1, quirk2 = random.sample(QUIRKS, 2)
|
||||
people_pool = PEOPLE_MALE if gender == "male" else PEOPLE_FEMALE
|
||||
person1, person2 = random.sample(people_pool, 2)
|
||||
tic1, tic2 = random.sample(VERBAL_TICS, 2)
|
||||
arc = random.choice(EMOTIONAL_ARCS)
|
||||
vehicle = random.choice(VEHICLES)
|
||||
having = random.choice(HAVING_RIGHT_NOW)
|
||||
|
||||
time_ctx = _get_time_context()
|
||||
moon = _get_moon_phase()
|
||||
season_ctx = _get_seasonal_context()
|
||||
|
||||
trait_str = ", ".join(traits) if traits else "a regular caller"
|
||||
|
||||
parts = [
|
||||
f"{age}, {job} {location}. Returning caller — {trait_str}.",
|
||||
f"{interest1.capitalize()}, {interest2}.",
|
||||
f"{quirk1.capitalize()}, {quirk2}.",
|
||||
f"\nRIGHT NOW: {time_ctx} Moon: {moon}.",
|
||||
f"\nSEASON: {season_ctx}",
|
||||
f"\nPEOPLE IN THEIR LIFE: {person1.capitalize()}. {person2.capitalize()}. Use their names when talking about them.",
|
||||
f"\nDRIVES: {vehicle.capitalize()}.",
|
||||
f"\nHAVING RIGHT NOW: {having}",
|
||||
f"\nVERBAL HABITS: Tends to say \"{tic1}\" and \"{tic2}\" — use these naturally in conversation.",
|
||||
f"\nEMOTIONAL ARC: {arc}",
|
||||
f"\nRELATIONSHIP TO THE SHOW: Has called before. Comfortable on air. Knows Luke a bit. Might reference their last call.",
|
||||
prev_section,
|
||||
]
|
||||
|
||||
return " ".join(parts[:3]) + "".join(parts[3:])
|
||||
|
||||
|
||||
def generate_caller_background(base: dict) -> str:
|
||||
"""Generate a unique background for a caller (sync, no research).
|
||||
~30% of callers are 'topic callers' who call about something interesting
|
||||
instead of a personal problem. Includes full personality layers for realism."""
|
||||
if base.get("returning") and base.get("regular_id"):
|
||||
return _generate_returning_caller_background(base)
|
||||
gender = base["gender"]
|
||||
age = random.randint(*base["age_range"])
|
||||
jobs = JOBS_MALE if gender == "male" else JOBS_FEMALE
|
||||
@@ -1491,8 +1579,58 @@ async def enrich_caller_background(background: str) -> str:
|
||||
|
||||
return background
|
||||
|
||||
def detect_host_mood(messages: list[dict]) -> str:
|
||||
"""Analyze recent host messages to detect mood signals for caller adaptation."""
|
||||
host_msgs = [m["content"] for m in messages if m.get("role") in ("user", "host")][-5:]
|
||||
if not host_msgs:
|
||||
return ""
|
||||
|
||||
signals = []
|
||||
|
||||
# Check average word count — short responses suggest dismissiveness
|
||||
avg_words = sum(len(m.split()) for m in host_msgs) / len(host_msgs)
|
||||
if avg_words < 8:
|
||||
signals.append("The host is giving short responses — they might be losing interest, testing you, or waiting for you to bring something real. Don't ramble. Get to the point or change the subject.")
|
||||
|
||||
# Pushback patterns
|
||||
pushback_phrases = ["i don't think", "that's not", "come on", "really?", "i disagree",
|
||||
"that doesn't", "are you sure", "i don't buy", "no way", "but that's",
|
||||
"hold on", "wait a minute", "let's be honest"]
|
||||
pushback_count = sum(1 for m in host_msgs for p in pushback_phrases if p in m.lower())
|
||||
if pushback_count >= 2:
|
||||
signals.append("The host is pushing back — they're challenging you. Don't fold immediately. Defend your position or concede specifically, not generically.")
|
||||
|
||||
# Supportive patterns
|
||||
supportive_phrases = ["i hear you", "that makes sense", "i get it", "that's real",
|
||||
"i feel you", "you're right", "absolutely", "exactly", "good for you",
|
||||
"i respect that", "that took guts", "i'm glad you"]
|
||||
supportive_count = sum(1 for m in host_msgs for p in supportive_phrases if p in m.lower())
|
||||
if supportive_count >= 2:
|
||||
signals.append("The host is being supportive — they're with you. You can go deeper. Share something you've been holding back.")
|
||||
|
||||
# Joking patterns
|
||||
joke_indicators = ["haha", "lmao", "lol", "that's hilarious", "no way", "you're killing me",
|
||||
"shut up", "get out", "are you serious", "you're joking"]
|
||||
joke_count = sum(1 for m in host_msgs for p in joke_indicators if p in m.lower())
|
||||
if joke_count >= 2:
|
||||
signals.append("The host is in a playful mood — joking around. You can joke back, lean into the humor, but you can also use it as a door to something real.")
|
||||
|
||||
# Probing — lots of questions
|
||||
question_count = sum(m.count("?") for m in host_msgs)
|
||||
if question_count >= 3:
|
||||
signals.append("The host is asking a lot of questions — they're digging. Give them real answers. Don't deflect.")
|
||||
|
||||
if not signals:
|
||||
return ""
|
||||
|
||||
# Cap at 2 signals
|
||||
signals = signals[:2]
|
||||
return "\nEMOTIONAL READ ON THE HOST:\n" + "\n".join(f"- {s}" for s in signals) + "\n"
|
||||
|
||||
|
||||
def get_caller_prompt(caller: dict, conversation_summary: str = "", show_history: str = "",
|
||||
news_context: str = "", research_context: str = "") -> str:
|
||||
news_context: str = "", research_context: str = "",
|
||||
emotional_read: str = "") -> str:
|
||||
"""Generate a natural system prompt for a caller"""
|
||||
context = ""
|
||||
if conversation_summary:
|
||||
@@ -1519,7 +1657,7 @@ Continue naturally. Don't repeat yourself.
|
||||
return f"""You're {caller['name']}, calling a late-night radio show called "Luke at the Roost." It's late. You trust this host.
|
||||
|
||||
{caller['vibe']}
|
||||
{history}{context}{world_context}
|
||||
{history}{context}{world_context}{emotional_read}
|
||||
HOW TO TALK:
|
||||
- Sound like a real person on the phone, not an essay. This is a conversation, not a monologue.
|
||||
- VARY YOUR LENGTH. Sometimes one sentence. Sometimes two or three. Match the moment.
|
||||
@@ -1607,6 +1745,8 @@ class CallRecord:
|
||||
caller_name: str # "Tony" or "Caller #3"
|
||||
summary: str # LLM-generated summary after hangup
|
||||
transcript: list[dict] = field(default_factory=list)
|
||||
started_at: float = 0.0
|
||||
ended_at: float = 0.0
|
||||
|
||||
|
||||
class Session:
|
||||
@@ -1616,6 +1756,7 @@ class Session:
|
||||
self.conversation: list[dict] = []
|
||||
self.caller_backgrounds: dict[str, str] = {} # Generated backgrounds for this session
|
||||
self.call_history: list[CallRecord] = []
|
||||
self._call_started_at: float = 0.0
|
||||
self.active_real_caller: dict | None = None
|
||||
self.ai_respond_mode: str = "manual" # "manual" or "auto"
|
||||
self.auto_followup: bool = False
|
||||
@@ -1626,13 +1767,14 @@ class Session:
|
||||
def start_call(self, caller_key: str):
|
||||
self.current_caller_key = caller_key
|
||||
self.conversation = []
|
||||
self._call_started_at = time.time()
|
||||
|
||||
def end_call(self):
|
||||
self.current_caller_key = None
|
||||
self.conversation = []
|
||||
|
||||
def add_message(self, role: str, content: str):
|
||||
self.conversation.append({"role": role, "content": content})
|
||||
self.conversation.append({"role": role, "content": content, "timestamp": time.time()})
|
||||
|
||||
def get_caller_background(self, caller_key: str) -> str:
|
||||
"""Get or generate background for a caller in this session"""
|
||||
@@ -1977,7 +2119,7 @@ async def get_callers():
|
||||
"""Get list of available callers"""
|
||||
return {
|
||||
"callers": [
|
||||
{"key": k, "name": v["name"]}
|
||||
{"key": k, "name": v["name"], "returning": v.get("returning", False)}
|
||||
for k, v in CALLER_BASES.items()
|
||||
],
|
||||
"current": session.current_caller_key,
|
||||
@@ -1985,6 +2127,12 @@ async def get_callers():
|
||||
}
|
||||
|
||||
|
||||
@app.get("/api/regulars")
|
||||
async def get_regulars():
|
||||
"""Get list of regular callers"""
|
||||
return {"regulars": regular_caller_service.get_regulars()}
|
||||
|
||||
|
||||
@app.post("/api/session/reset")
|
||||
async def reset_session():
|
||||
"""Reset session - all callers get fresh backgrounds"""
|
||||
@@ -2037,6 +2185,9 @@ async def hangup():
|
||||
session._research_task = None
|
||||
|
||||
caller_name = session.caller["name"] if session.caller else None
|
||||
caller_key = session.current_caller_key
|
||||
conversation_snapshot = list(session.conversation)
|
||||
call_started = getattr(session, '_call_started_at', 0.0)
|
||||
session.end_call()
|
||||
|
||||
# Play hangup sound in background so response returns immediately
|
||||
@@ -2044,9 +2195,74 @@ async def hangup():
|
||||
if hangup_sound.exists():
|
||||
threading.Thread(target=audio_service.play_sfx, args=(str(hangup_sound),), daemon=True).start()
|
||||
|
||||
# Generate summary for AI caller in background
|
||||
if caller_name and conversation_snapshot:
|
||||
asyncio.create_task(_summarize_ai_call(caller_key, caller_name, conversation_snapshot, call_started))
|
||||
|
||||
return {"status": "disconnected", "caller": caller_name}
|
||||
|
||||
|
||||
async def _summarize_ai_call(caller_key: str, caller_name: str, conversation: list[dict], started_at: float):
|
||||
"""Background task: summarize AI caller conversation and store in history"""
|
||||
ended_at = time.time()
|
||||
summary = ""
|
||||
if conversation:
|
||||
transcript_text = "\n".join(
|
||||
f"{msg['role']}: {msg['content']}" for msg in conversation
|
||||
)
|
||||
try:
|
||||
summary = await llm_service.generate(
|
||||
messages=[{"role": "user", "content": f"Summarize this radio show call in 1-2 sentences:\n{transcript_text}"}],
|
||||
system_prompt="You summarize radio show conversations concisely. Focus on what the caller talked about and any emotional moments.",
|
||||
)
|
||||
except Exception as e:
|
||||
print(f"[AI Summary] Failed to generate summary: {e}")
|
||||
summary = f"{caller_name} called in."
|
||||
|
||||
session.call_history.append(CallRecord(
|
||||
caller_type="ai",
|
||||
caller_name=caller_name,
|
||||
summary=summary,
|
||||
transcript=conversation,
|
||||
started_at=started_at,
|
||||
ended_at=ended_at,
|
||||
))
|
||||
print(f"[AI Summary] {caller_name} call summarized: {summary[:80]}...")
|
||||
|
||||
# Returning caller promotion/update logic
|
||||
try:
|
||||
base = CALLER_BASES.get(caller_key) if caller_key else None
|
||||
if base and summary:
|
||||
if base.get("returning") and base.get("regular_id"):
|
||||
# Update existing regular's call history
|
||||
regular_caller_service.update_after_call(base["regular_id"], summary)
|
||||
elif len(conversation) >= 6 and random.random() < 0.20:
|
||||
# 20% chance to promote first-timer with 6+ messages
|
||||
bg = session.caller_backgrounds.get(caller_key, "")
|
||||
traits = []
|
||||
for label in ["QUIRK", "STRONG OPINION", "SECRET SIDE", "FOOD OPINION"]:
|
||||
for line in bg.split("\n"):
|
||||
if label in line:
|
||||
traits.append(line.split(":", 1)[-1].strip()[:80])
|
||||
break
|
||||
# Extract job and location from first line of background
|
||||
first_line = bg.split(".")[0] if bg else ""
|
||||
parts = first_line.split(",", 1)
|
||||
job_loc = parts[1].strip() if len(parts) > 1 else ""
|
||||
job_parts = job_loc.rsplit(" in ", 1) if " in " in job_loc else (job_loc, "unknown")
|
||||
regular_caller_service.add_regular(
|
||||
name=caller_name,
|
||||
gender=base.get("gender", "male"),
|
||||
age=random.randint(*base.get("age_range", (30, 50))),
|
||||
job=job_parts[0].strip() if isinstance(job_parts, tuple) else job_parts[0],
|
||||
location="in " + job_parts[1].strip() if isinstance(job_parts, tuple) and len(job_parts) > 1 else "unknown",
|
||||
personality_traits=traits[:4],
|
||||
first_call_summary=summary,
|
||||
)
|
||||
except Exception as e:
|
||||
print(f"[Regulars] Promotion logic error: {e}")
|
||||
|
||||
|
||||
# --- Chat & TTS Endpoints ---
|
||||
|
||||
import re
|
||||
@@ -2174,7 +2390,8 @@ async def chat(request: ChatRequest):
|
||||
|
||||
conversation_summary = session.get_conversation_summary()
|
||||
show_history = session.get_show_history()
|
||||
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history)
|
||||
mood = detect_host_mood(session.conversation)
|
||||
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history, emotional_read=mood)
|
||||
|
||||
messages = _normalize_messages_for_llm(session.conversation[-10:])
|
||||
response = await llm_service.generate(
|
||||
@@ -2276,11 +2493,14 @@ async def get_music():
|
||||
|
||||
@app.post("/api/music/play")
|
||||
async def play_music(request: MusicRequest):
|
||||
"""Load and play a music track"""
|
||||
"""Load and play a music track, crossfading if already playing"""
|
||||
track_path = settings.music_dir / request.track
|
||||
if not track_path.exists():
|
||||
raise HTTPException(404, "Track not found")
|
||||
|
||||
if audio_service.is_music_playing():
|
||||
audio_service.crossfade_to(str(track_path))
|
||||
else:
|
||||
audio_service.load_music(str(track_path))
|
||||
audio_service.play_music()
|
||||
return {"status": "playing", "track": request.track}
|
||||
@@ -2352,6 +2572,9 @@ async def play_ad(request: MusicRequest):
|
||||
if not ad_path.exists():
|
||||
raise HTTPException(404, "Ad not found")
|
||||
|
||||
if audio_service._music_playing:
|
||||
audio_service.stop_music(fade_duration=1.0)
|
||||
await asyncio.sleep(1.1)
|
||||
audio_service.play_ad(str(ad_path))
|
||||
return {"status": "playing", "track": request.track}
|
||||
|
||||
@@ -2393,6 +2616,126 @@ async def update_settings(data: dict):
|
||||
return llm_service.get_settings()
|
||||
|
||||
|
||||
# --- Caller Screening ---
|
||||
|
||||
SCREENING_PROMPT = """You are a friendly, brief phone screener for "Luke at the Roost" radio show.
|
||||
Your job: Get the caller's first name and what they want to talk about. That's it.
|
||||
|
||||
Rules:
|
||||
- Be warm but brief (1-2 sentences per response)
|
||||
- First ask their name, then ask what they want to talk about
|
||||
- After you have both, say something like "Great, sit tight and we'll get you on with Luke!"
|
||||
- Never pretend to be Luke or the host
|
||||
- Keep it casual and conversational
|
||||
- If they're hard to understand, ask them to repeat"""
|
||||
|
||||
_screening_audio_buffers: dict[str, bytearray] = {}
|
||||
|
||||
|
||||
async def _start_screening_greeting(caller_id: str):
|
||||
"""Send initial screening greeting to queued caller after brief delay"""
|
||||
await asyncio.sleep(2) # Wait for stream to stabilize
|
||||
|
||||
ws = caller_service._websockets.get(caller_id)
|
||||
if not ws:
|
||||
return
|
||||
|
||||
caller_service.start_screening(caller_id)
|
||||
greeting = "Hey there! Thanks for calling Luke at the Roost. What's your name?"
|
||||
caller_service.update_screening(caller_id, screener_text=greeting)
|
||||
|
||||
try:
|
||||
audio_bytes = await generate_speech(greeting, "Sarah", "none")
|
||||
if audio_bytes:
|
||||
await caller_service.stream_audio_to_caller(caller_id, audio_bytes, 24000)
|
||||
except Exception as e:
|
||||
print(f"[Screening] Greeting TTS failed: {e}")
|
||||
|
||||
|
||||
async def _handle_screening_audio(caller_id: str, pcm_data: bytes, sample_rate: int):
|
||||
"""Process audio from a queued caller for screening conversation"""
|
||||
state = caller_service.get_screening_state(caller_id)
|
||||
if not state or state["status"] == "complete":
|
||||
return
|
||||
|
||||
# Skip if TTS is currently streaming to this caller
|
||||
if caller_service.is_streaming_tts(caller_id):
|
||||
return
|
||||
|
||||
# Transcribe caller speech
|
||||
try:
|
||||
text = await transcribe_audio(pcm_data, source_sample_rate=sample_rate)
|
||||
except Exception as e:
|
||||
print(f"[Screening] Transcription failed: {e}")
|
||||
return
|
||||
|
||||
if not text or not text.strip():
|
||||
return
|
||||
|
||||
print(f"[Screening] Caller {caller_id}: {text}")
|
||||
caller_service.update_screening(caller_id, caller_text=text)
|
||||
|
||||
# Build conversation for LLM
|
||||
messages = []
|
||||
for msg in state["conversation"]:
|
||||
role = "assistant" if msg["role"] == "screener" else "user"
|
||||
messages.append({"role": role, "content": msg["content"]})
|
||||
|
||||
# Generate screener response
|
||||
try:
|
||||
response = await llm_service.generate(
|
||||
messages=messages,
|
||||
system_prompt=SCREENING_PROMPT
|
||||
)
|
||||
except Exception as e:
|
||||
print(f"[Screening] LLM failed: {e}")
|
||||
return
|
||||
|
||||
if not response or not response.strip():
|
||||
return
|
||||
|
||||
response = response.strip()
|
||||
print(f"[Screening] Screener → {caller_id}: {response}")
|
||||
caller_service.update_screening(caller_id, screener_text=response)
|
||||
|
||||
# After 2+ caller responses, try to extract name and topic
|
||||
if state["response_count"] >= 2:
|
||||
try:
|
||||
extract_prompt = f"""From this screening conversation, extract the caller's name and topic.
|
||||
Conversation:
|
||||
{chr(10).join(f'{m["role"]}: {m["content"]}' for m in state["conversation"])}
|
||||
|
||||
Respond with ONLY JSON: {{"name": "their first name or null", "topic": "brief topic or null"}}"""
|
||||
extract = await llm_service.generate(
|
||||
messages=[{"role": "user", "content": extract_prompt}],
|
||||
system_prompt="You extract structured data from conversations. Respond with only valid JSON."
|
||||
)
|
||||
json_match = re.search(r'\{[^}]+\}', extract)
|
||||
if json_match:
|
||||
info = json.loads(json_match.group())
|
||||
if info.get("name"):
|
||||
caller_service.update_screening(caller_id, caller_name=info["name"])
|
||||
if info.get("topic"):
|
||||
caller_service.update_screening(caller_id, topic=info["topic"])
|
||||
if info.get("name") and info.get("topic"):
|
||||
caller_service.end_screening(caller_id)
|
||||
broadcast_event("screening_complete", {
|
||||
"caller_id": caller_id,
|
||||
"name": info["name"],
|
||||
"topic": info["topic"]
|
||||
})
|
||||
except Exception as e:
|
||||
print(f"[Screening] Extract failed: {e}")
|
||||
|
||||
# TTS the screener response back to caller
|
||||
try:
|
||||
audio_bytes = await generate_speech(response, "Sarah", "none")
|
||||
if audio_bytes:
|
||||
await caller_service.stream_audio_to_caller(caller_id, audio_bytes, 24000)
|
||||
except Exception as e:
|
||||
print(f"[Screening] Response TTS failed: {e}")
|
||||
|
||||
|
||||
@app.websocket("/api/signalwire/stream")
|
||||
async def signalwire_audio_stream(websocket: WebSocket):
|
||||
"""Handle SignalWire bidirectional audio stream"""
|
||||
@@ -2402,6 +2745,7 @@ async def signalwire_audio_stream(websocket: WebSocket):
|
||||
caller_phone = "Unknown"
|
||||
call_sid = ""
|
||||
audio_buffer = bytearray()
|
||||
screening_buffer = bytearray()
|
||||
CHUNK_DURATION_S = 3
|
||||
SAMPLE_RATE = 16000
|
||||
chunk_samples = CHUNK_DURATION_S * SAMPLE_RATE
|
||||
@@ -2448,6 +2792,9 @@ async def signalwire_audio_stream(websocket: WebSocket):
|
||||
if stream_sid:
|
||||
caller_service.register_stream_sid(caller_id, stream_sid)
|
||||
|
||||
# Start screening conversation
|
||||
asyncio.create_task(_start_screening_greeting(caller_id))
|
||||
|
||||
elif event == "media" and stream_started:
|
||||
try:
|
||||
payload = msg.get("media", {}).get("payload", "")
|
||||
@@ -2458,6 +2805,16 @@ async def signalwire_audio_stream(websocket: WebSocket):
|
||||
|
||||
call_info = caller_service.active_calls.get(caller_id)
|
||||
if not call_info:
|
||||
# Caller is queued, not on air — route to screening
|
||||
screening_buffer.extend(pcm_data)
|
||||
if len(screening_buffer) >= chunk_samples * 2:
|
||||
pcm_chunk = bytes(screening_buffer[:chunk_samples * 2])
|
||||
screening_buffer = screening_buffer[chunk_samples * 2:]
|
||||
audio_check = np.frombuffer(pcm_chunk, dtype=np.int16).astype(np.float32) / 32768.0
|
||||
if np.abs(audio_check).max() >= 0.01:
|
||||
asyncio.create_task(
|
||||
_handle_screening_audio(caller_id, pcm_chunk, SAMPLE_RATE)
|
||||
)
|
||||
continue
|
||||
|
||||
audio_buffer.extend(pcm_data)
|
||||
@@ -2713,7 +3070,8 @@ async def _trigger_ai_auto_respond(accumulated_text: str):
|
||||
|
||||
conversation_summary = session.get_conversation_summary()
|
||||
show_history = session.get_show_history()
|
||||
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history)
|
||||
mood = detect_host_mood(session.conversation)
|
||||
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history, emotional_read=mood)
|
||||
|
||||
messages = _normalize_messages_for_llm(session.conversation[-10:])
|
||||
response = await llm_service.generate(
|
||||
@@ -2785,7 +3143,8 @@ async def ai_respond():
|
||||
|
||||
conversation_summary = session.get_conversation_summary()
|
||||
show_history = session.get_show_history()
|
||||
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history)
|
||||
mood = detect_host_mood(session.conversation)
|
||||
system_prompt = get_caller_prompt(session.caller, conversation_summary, show_history, emotional_read=mood)
|
||||
|
||||
messages = _normalize_messages_for_llm(session.conversation[-10:])
|
||||
response = await llm_service.generate(
|
||||
@@ -2856,6 +3215,7 @@ async def hangup_real_caller():
|
||||
caller_id = session.active_real_caller["caller_id"]
|
||||
caller_phone = session.active_real_caller["phone"]
|
||||
conversation_snapshot = list(session.conversation)
|
||||
call_started = getattr(session, '_call_started_at', 0.0)
|
||||
auto_followup_enabled = session.auto_followup
|
||||
|
||||
# End the phone call via SignalWire
|
||||
@@ -2875,7 +3235,7 @@ async def hangup_real_caller():
|
||||
threading.Thread(target=audio_service.play_sfx, args=(str(hangup_sound),), daemon=True).start()
|
||||
|
||||
asyncio.create_task(
|
||||
_summarize_real_call(caller_phone, conversation_snapshot, auto_followup_enabled)
|
||||
_summarize_real_call(caller_phone, conversation_snapshot, call_started, auto_followup_enabled)
|
||||
)
|
||||
|
||||
return {
|
||||
@@ -2884,8 +3244,9 @@ async def hangup_real_caller():
|
||||
}
|
||||
|
||||
|
||||
async def _summarize_real_call(caller_phone: str, conversation: list, auto_followup_enabled: bool):
|
||||
async def _summarize_real_call(caller_phone: str, conversation: list, started_at: float, auto_followup_enabled: bool):
|
||||
"""Background task: summarize call and store in history"""
|
||||
ended_at = time.time()
|
||||
summary = ""
|
||||
if conversation:
|
||||
transcript_text = "\n".join(
|
||||
@@ -2901,6 +3262,8 @@ async def _summarize_real_call(caller_phone: str, conversation: list, auto_follo
|
||||
caller_name=caller_phone,
|
||||
summary=summary,
|
||||
transcript=conversation,
|
||||
started_at=started_at,
|
||||
ended_at=ended_at,
|
||||
))
|
||||
print(f"[Real Caller] {caller_phone} call summarized: {summary[:80]}...")
|
||||
|
||||
@@ -2963,6 +3326,70 @@ async def set_auto_followup(data: dict):
|
||||
return {"enabled": session.auto_followup}
|
||||
|
||||
|
||||
# --- Transcript & Chapter Export ---
|
||||
|
||||
@app.get("/api/session/export")
|
||||
async def export_session():
|
||||
"""Export session transcript with speaker labels and chapters from call boundaries"""
|
||||
if not session.call_history:
|
||||
raise HTTPException(400, "No calls in this session to export")
|
||||
|
||||
# Find the earliest call start as session base time
|
||||
session_start = min(
|
||||
(r.started_at for r in session.call_history if r.started_at > 0),
|
||||
default=time.time()
|
||||
)
|
||||
|
||||
transcript_lines = []
|
||||
chapters = []
|
||||
|
||||
for i, record in enumerate(session.call_history):
|
||||
# Chapter from call start time
|
||||
offset_seconds = max(0, record.started_at - session_start) if record.started_at > 0 else 0
|
||||
chapter_title = f"{record.caller_name}"
|
||||
if record.summary:
|
||||
# Use first sentence of summary for chapter title
|
||||
short_summary = record.summary.split(".")[0].strip()
|
||||
if short_summary:
|
||||
chapter_title += f" \u2014 {short_summary}"
|
||||
chapters.append({"startTime": round(offset_seconds), "title": chapter_title})
|
||||
|
||||
# Separator between calls
|
||||
if i > 0:
|
||||
transcript_lines.append("")
|
||||
transcript_lines.append(f"--- Call {i + 1}: {record.caller_name} ---")
|
||||
transcript_lines.append("")
|
||||
|
||||
# Transcript lines with timestamps
|
||||
for msg in record.transcript:
|
||||
msg_offset = msg.get("timestamp", 0) - session_start if msg.get("timestamp") else offset_seconds
|
||||
if msg_offset < 0:
|
||||
msg_offset = 0
|
||||
mins = int(msg_offset // 60)
|
||||
secs = int(msg_offset % 60)
|
||||
|
||||
role = msg.get("role", "")
|
||||
if role in ("user", "host"):
|
||||
speaker = "HOST"
|
||||
elif role.startswith("real_caller:"):
|
||||
speaker = role.split(":", 1)[1].upper()
|
||||
elif role.startswith("ai_caller:"):
|
||||
speaker = role.split(":", 1)[1].upper()
|
||||
elif role == "assistant":
|
||||
speaker = record.caller_name.upper()
|
||||
else:
|
||||
speaker = role.upper()
|
||||
|
||||
transcript_lines.append(f"[{mins:02d}:{secs:02d}] {speaker}: {msg['content']}")
|
||||
|
||||
return {
|
||||
"session_id": session.id,
|
||||
"transcript": "\n".join(transcript_lines),
|
||||
"chapters": chapters,
|
||||
"call_count": len(session.call_history),
|
||||
}
|
||||
|
||||
|
||||
# --- Server Control Endpoints ---
|
||||
|
||||
import subprocess
|
||||
|
||||
@@ -53,6 +53,14 @@ class AudioService:
|
||||
self._music_volume: float = 0.3
|
||||
self._music_loop: bool = True
|
||||
|
||||
# Music crossfade state
|
||||
self._crossfade_active: bool = False
|
||||
self._crossfade_old_data: Optional[np.ndarray] = None
|
||||
self._crossfade_old_position: int = 0
|
||||
self._crossfade_progress: float = 0.0
|
||||
self._crossfade_samples: int = 0
|
||||
self._crossfade_step: float = 0.0
|
||||
|
||||
# Caller playback state
|
||||
self._caller_stop_event = threading.Event()
|
||||
self._caller_thread: Optional[threading.Thread] = None
|
||||
@@ -578,6 +586,55 @@ class AudioService:
|
||||
print(f"Failed to load music: {e}")
|
||||
return False
|
||||
|
||||
def crossfade_to(self, file_path: str, duration: float = 3.0):
|
||||
"""Crossfade from current music track to a new one"""
|
||||
import librosa
|
||||
|
||||
if not self._music_playing or self._music_resampled is None:
|
||||
if self.load_music(file_path):
|
||||
self.play_music()
|
||||
return
|
||||
|
||||
# Load the new track
|
||||
path = Path(file_path)
|
||||
if not path.exists():
|
||||
print(f"Music file not found: {file_path}")
|
||||
return
|
||||
|
||||
try:
|
||||
audio, sr = librosa.load(str(path), sr=self.output_sample_rate, mono=True)
|
||||
new_data = audio.astype(np.float32)
|
||||
except Exception as e:
|
||||
print(f"Failed to load music for crossfade: {e}")
|
||||
return
|
||||
|
||||
# Get device sample rate for resampling
|
||||
if self.output_device is not None:
|
||||
device_info = sd.query_devices(self.output_device)
|
||||
device_sr = int(device_info['default_samplerate'])
|
||||
else:
|
||||
device_sr = self.output_sample_rate
|
||||
|
||||
if self.output_sample_rate != device_sr:
|
||||
new_resampled = librosa.resample(new_data, orig_sr=self.output_sample_rate, target_sr=device_sr)
|
||||
else:
|
||||
new_resampled = new_data.copy()
|
||||
|
||||
# Swap: current becomes old, new becomes current
|
||||
self._crossfade_old_data = self._music_resampled
|
||||
self._crossfade_old_position = self._music_position
|
||||
self._music_resampled = new_resampled
|
||||
self._music_data = new_data
|
||||
self._music_position = 0
|
||||
|
||||
# Configure crossfade timing
|
||||
self._crossfade_samples = int(device_sr * duration)
|
||||
self._crossfade_progress = 0.0
|
||||
self._crossfade_step = 1.0 / self._crossfade_samples if self._crossfade_samples > 0 else 1.0
|
||||
self._crossfade_active = True
|
||||
|
||||
print(f"Crossfading to {path.name} over {duration}s")
|
||||
|
||||
def play_music(self):
|
||||
"""Start music playback to specific channel"""
|
||||
import librosa
|
||||
@@ -625,25 +682,55 @@ class AudioService:
|
||||
if not self._music_playing or self._music_resampled is None:
|
||||
return
|
||||
|
||||
# Read new track samples
|
||||
end_pos = self._music_position + frames
|
||||
|
||||
if end_pos <= len(self._music_resampled):
|
||||
outdata[:, channel_idx] = self._music_resampled[self._music_position:end_pos] * self._music_volume
|
||||
new_samples = self._music_resampled[self._music_position:end_pos].copy()
|
||||
self._music_position = end_pos
|
||||
else:
|
||||
remaining = len(self._music_resampled) - self._music_position
|
||||
new_samples = np.zeros(frames, dtype=np.float32)
|
||||
if remaining > 0:
|
||||
outdata[:remaining, channel_idx] = self._music_resampled[self._music_position:] * self._music_volume
|
||||
|
||||
new_samples[:remaining] = self._music_resampled[self._music_position:]
|
||||
if self._music_loop:
|
||||
self._music_position = 0
|
||||
wrap_frames = frames - remaining
|
||||
if wrap_frames > 0:
|
||||
outdata[remaining:, channel_idx] = self._music_resampled[:wrap_frames] * self._music_volume
|
||||
new_samples[remaining:] = self._music_resampled[:wrap_frames]
|
||||
self._music_position = wrap_frames
|
||||
else:
|
||||
self._music_position = len(self._music_resampled)
|
||||
if remaining <= 0:
|
||||
self._music_playing = False
|
||||
|
||||
if self._crossfade_active and self._crossfade_old_data is not None:
|
||||
# Read old track samples
|
||||
old_end = self._crossfade_old_position + frames
|
||||
if old_end <= len(self._crossfade_old_data):
|
||||
old_samples = self._crossfade_old_data[self._crossfade_old_position:old_end]
|
||||
self._crossfade_old_position = old_end
|
||||
else:
|
||||
old_remaining = len(self._crossfade_old_data) - self._crossfade_old_position
|
||||
old_samples = np.zeros(frames, dtype=np.float32)
|
||||
if old_remaining > 0:
|
||||
old_samples[:old_remaining] = self._crossfade_old_data[self._crossfade_old_position:]
|
||||
self._crossfade_old_position = len(self._crossfade_old_data)
|
||||
|
||||
# Compute fade curves for this chunk
|
||||
start_progress = self._crossfade_progress
|
||||
end_progress = min(1.0, start_progress + self._crossfade_step * frames)
|
||||
fade_in = np.linspace(start_progress, end_progress, frames, dtype=np.float32)
|
||||
fade_out = 1.0 - fade_in
|
||||
|
||||
outdata[:, channel_idx] = (old_samples * fade_out + new_samples * fade_in) * self._music_volume
|
||||
self._crossfade_progress = end_progress
|
||||
|
||||
if self._crossfade_progress >= 1.0:
|
||||
self._crossfade_active = False
|
||||
self._crossfade_old_data = None
|
||||
print("Crossfade complete")
|
||||
else:
|
||||
outdata[:, channel_idx] = new_samples * self._music_volume
|
||||
|
||||
try:
|
||||
self._music_stream = sd.OutputStream(
|
||||
device=device,
|
||||
@@ -659,15 +746,48 @@ class AudioService:
|
||||
print(f"Music playback error: {e}")
|
||||
self._music_playing = False
|
||||
|
||||
def stop_music(self):
|
||||
"""Stop music playback"""
|
||||
def stop_music(self, fade_duration: float = 2.0):
|
||||
"""Stop music playback with fade out"""
|
||||
if not self._music_playing or not self._music_stream:
|
||||
self._music_playing = False
|
||||
if self._music_stream:
|
||||
self._music_stream.stop()
|
||||
self._music_stream.close()
|
||||
self._music_stream = None
|
||||
self._music_position = 0
|
||||
return
|
||||
|
||||
if fade_duration <= 0:
|
||||
self._music_playing = False
|
||||
self._music_stream.stop()
|
||||
self._music_stream.close()
|
||||
self._music_stream = None
|
||||
self._music_position = 0
|
||||
print("Music stopped")
|
||||
return
|
||||
|
||||
import threading
|
||||
original_volume = self._music_volume
|
||||
steps = 20
|
||||
step_time = fade_duration / steps
|
||||
|
||||
def _fade():
|
||||
for i in range(steps):
|
||||
if not self._music_playing:
|
||||
break
|
||||
self._music_volume = original_volume * (1 - (i + 1) / steps)
|
||||
import time
|
||||
time.sleep(step_time)
|
||||
self._music_playing = False
|
||||
if self._music_stream:
|
||||
self._music_stream.stop()
|
||||
self._music_stream.close()
|
||||
self._music_stream = None
|
||||
self._music_position = 0
|
||||
self._music_volume = original_volume
|
||||
print("Music faded out and stopped")
|
||||
|
||||
threading.Thread(target=_fade, daemon=True).start()
|
||||
|
||||
def play_ad(self, file_path: str):
|
||||
"""Load and play an ad file once (no loop) on the ad channel"""
|
||||
|
||||
@@ -25,6 +25,7 @@ class CallerService:
|
||||
self._stream_sids: dict[str, str] = {} # caller_id -> SignalWire streamSid
|
||||
self._send_locks: dict[str, asyncio.Lock] = {} # per-caller send lock
|
||||
self._streaming_tts: set[str] = set() # caller_ids currently receiving TTS
|
||||
self._screening_state: dict[str, dict] = {} # caller_id -> screening conversation
|
||||
|
||||
def _get_send_lock(self, caller_id: str) -> asyncio.Lock:
|
||||
if caller_id not in self._send_locks:
|
||||
@@ -51,18 +52,6 @@ class CallerService:
|
||||
self._queue = [c for c in self._queue if c["caller_id"] != caller_id]
|
||||
print(f"[Caller] {caller_id} removed from queue")
|
||||
|
||||
def get_queue(self) -> list[dict]:
|
||||
now = time.time()
|
||||
with self._lock:
|
||||
return [
|
||||
{
|
||||
"caller_id": c["caller_id"],
|
||||
"phone": c["phone"],
|
||||
"wait_time": int(now - c["queued_at"]),
|
||||
}
|
||||
for c in self._queue
|
||||
]
|
||||
|
||||
def allocate_channel(self) -> int:
|
||||
with self._lock:
|
||||
ch = self.FIRST_REAL_CHANNEL
|
||||
@@ -111,6 +100,7 @@ class CallerService:
|
||||
self._call_sids.pop(caller_id, None)
|
||||
self._stream_sids.pop(caller_id, None)
|
||||
self._send_locks.pop(caller_id, None)
|
||||
self._screening_state.pop(caller_id, None)
|
||||
|
||||
def reset(self):
|
||||
with self._lock:
|
||||
@@ -125,8 +115,72 @@ class CallerService:
|
||||
self._stream_sids.clear()
|
||||
self._send_locks.clear()
|
||||
self._streaming_tts.clear()
|
||||
self._screening_state.clear()
|
||||
print("[Caller] Service reset")
|
||||
|
||||
# --- Screening ---
|
||||
|
||||
def start_screening(self, caller_id: str):
|
||||
"""Initialize screening state for a queued caller"""
|
||||
self._screening_state[caller_id] = {
|
||||
"conversation": [],
|
||||
"caller_name": None,
|
||||
"topic": None,
|
||||
"status": "screening", # screening, complete
|
||||
"response_count": 0,
|
||||
}
|
||||
print(f"[Screening] Started for {caller_id}")
|
||||
|
||||
def get_screening_state(self, caller_id: str) -> Optional[dict]:
|
||||
return self._screening_state.get(caller_id)
|
||||
|
||||
def update_screening(self, caller_id: str, caller_text: str = None,
|
||||
screener_text: str = None, caller_name: str = None,
|
||||
topic: str = None):
|
||||
"""Update screening conversation and extracted info"""
|
||||
state = self._screening_state.get(caller_id)
|
||||
if not state:
|
||||
return
|
||||
if caller_text:
|
||||
state["conversation"].append({"role": "caller", "content": caller_text})
|
||||
state["response_count"] += 1
|
||||
if screener_text:
|
||||
state["conversation"].append({"role": "screener", "content": screener_text})
|
||||
if caller_name:
|
||||
state["caller_name"] = caller_name
|
||||
if topic:
|
||||
state["topic"] = topic
|
||||
|
||||
def end_screening(self, caller_id: str):
|
||||
"""Mark screening as complete"""
|
||||
state = self._screening_state.get(caller_id)
|
||||
if state:
|
||||
state["status"] = "complete"
|
||||
print(f"[Screening] Complete for {caller_id}: name={state.get('caller_name')}, topic={state.get('topic')}")
|
||||
|
||||
def get_queue(self) -> list[dict]:
|
||||
"""Get queue with screening info enrichment"""
|
||||
now = time.time()
|
||||
with self._lock:
|
||||
result = []
|
||||
for c in self._queue:
|
||||
entry = {
|
||||
"caller_id": c["caller_id"],
|
||||
"phone": c["phone"],
|
||||
"wait_time": int(now - c["queued_at"]),
|
||||
}
|
||||
screening = self._screening_state.get(c["caller_id"])
|
||||
if screening:
|
||||
entry["screening_status"] = screening["status"]
|
||||
entry["caller_name"] = screening.get("caller_name")
|
||||
entry["screening_summary"] = screening.get("topic")
|
||||
else:
|
||||
entry["screening_status"] = None
|
||||
entry["caller_name"] = None
|
||||
entry["screening_summary"] = None
|
||||
result.append(entry)
|
||||
return result
|
||||
|
||||
def register_websocket(self, caller_id: str, websocket):
|
||||
"""Register a WebSocket for a caller"""
|
||||
self._websockets[caller_id] = websocket
|
||||
|
||||
95
backend/services/regulars.py
Normal file
95
backend/services/regulars.py
Normal file
@@ -0,0 +1,95 @@
|
||||
"""Returning caller persistence service"""
|
||||
|
||||
import json
|
||||
import time
|
||||
import uuid
|
||||
from pathlib import Path
|
||||
from typing import Optional
|
||||
|
||||
DATA_FILE = Path(__file__).parent.parent.parent / "data" / "regulars.json"
|
||||
MAX_REGULARS = 12
|
||||
|
||||
|
||||
class RegularCallerService:
|
||||
"""Manages persistent 'regular' callers who return across sessions"""
|
||||
|
||||
def __init__(self):
|
||||
self._regulars: list[dict] = []
|
||||
self._load()
|
||||
|
||||
def _load(self):
|
||||
if DATA_FILE.exists():
|
||||
try:
|
||||
with open(DATA_FILE) as f:
|
||||
data = json.load(f)
|
||||
self._regulars = data.get("regulars", [])
|
||||
print(f"[Regulars] Loaded {len(self._regulars)} regular callers")
|
||||
except Exception as e:
|
||||
print(f"[Regulars] Failed to load: {e}")
|
||||
self._regulars = []
|
||||
|
||||
def _save(self):
|
||||
try:
|
||||
DATA_FILE.parent.mkdir(parents=True, exist_ok=True)
|
||||
with open(DATA_FILE, "w") as f:
|
||||
json.dump({"regulars": self._regulars}, f, indent=2)
|
||||
except Exception as e:
|
||||
print(f"[Regulars] Failed to save: {e}")
|
||||
|
||||
def get_regulars(self) -> list[dict]:
|
||||
return list(self._regulars)
|
||||
|
||||
def get_returning_callers(self, count: int = 2) -> list[dict]:
|
||||
"""Get up to `count` regulars for returning caller slots"""
|
||||
import random
|
||||
if not self._regulars:
|
||||
return []
|
||||
available = [r for r in self._regulars if len(r.get("call_history", [])) > 0]
|
||||
if not available:
|
||||
return []
|
||||
return random.sample(available, min(count, len(available)))
|
||||
|
||||
def add_regular(self, name: str, gender: str, age: int, job: str,
|
||||
location: str, personality_traits: list[str],
|
||||
first_call_summary: str) -> dict:
|
||||
"""Promote a first-time caller to regular"""
|
||||
# Retire oldest if at cap
|
||||
if len(self._regulars) >= MAX_REGULARS:
|
||||
self._regulars.sort(key=lambda r: r.get("last_call", 0))
|
||||
retired = self._regulars.pop(0)
|
||||
print(f"[Regulars] Retired {retired['name']} to make room")
|
||||
|
||||
regular = {
|
||||
"id": str(uuid.uuid4())[:8],
|
||||
"name": name,
|
||||
"gender": gender,
|
||||
"age": age,
|
||||
"job": job,
|
||||
"location": location,
|
||||
"personality_traits": personality_traits,
|
||||
"call_history": [
|
||||
{"summary": first_call_summary, "timestamp": time.time()}
|
||||
],
|
||||
"last_call": time.time(),
|
||||
"created_at": time.time(),
|
||||
}
|
||||
self._regulars.append(regular)
|
||||
self._save()
|
||||
print(f"[Regulars] Promoted {name} to regular (total: {len(self._regulars)})")
|
||||
return regular
|
||||
|
||||
def update_after_call(self, regular_id: str, call_summary: str):
|
||||
"""Update a regular's history after a returning call"""
|
||||
for regular in self._regulars:
|
||||
if regular["id"] == regular_id:
|
||||
regular.setdefault("call_history", []).append(
|
||||
{"summary": call_summary, "timestamp": time.time()}
|
||||
)
|
||||
regular["last_call"] = time.time()
|
||||
self._save()
|
||||
print(f"[Regulars] Updated {regular['name']} call history ({len(regular['call_history'])} calls)")
|
||||
return
|
||||
print(f"[Regulars] Regular {regular_id} not found for update")
|
||||
|
||||
|
||||
regular_caller_service = RegularCallerService()
|
||||
1
data/regulars.json
Normal file
1
data/regulars.json
Normal file
@@ -0,0 +1 @@
|
||||
{"regulars": []}
|
||||
189
docs/plans/2026-02-05-real-callers-design.md
Normal file
189
docs/plans/2026-02-05-real-callers-design.md
Normal file
@@ -0,0 +1,189 @@
|
||||
# Real Callers + AI Follow-Up Design
|
||||
|
||||
## Overview
|
||||
|
||||
Add real phone callers to the AI Radio Show via Twilio, alongside existing AI callers. Real callers dial a phone number, wait in a hold queue, and get taken on air by the host. Three-way conversations between host, real caller, and AI caller are supported. AI follow-up callers automatically reference what real callers said.
|
||||
|
||||
## Requirements
|
||||
|
||||
- Real callers connect via Twilio phone number
|
||||
- Full-duplex audio — host and caller talk simultaneously, talk over each other
|
||||
- Each real caller gets their own dedicated audio channel for recording
|
||||
- Three-way calls: host + real caller + AI caller all live at once
|
||||
- AI caller can respond manually (host-triggered) or automatically (listens and decides when to jump in)
|
||||
- AI follow-up callers reference real caller conversations via show history
|
||||
- Auto follow-up mode: system picks an AI caller and connects them after a real call
|
||||
- Simple hold queue — callers wait with hold music, host sees list and picks who goes on air
|
||||
- Twilio webhooks exposed via Cloudflare tunnel
|
||||
|
||||
## Architecture
|
||||
|
||||
### Audio Routing (Loopback Channels)
|
||||
|
||||
```
|
||||
Ch 1: Host mic (existing)
|
||||
Ch 2: AI callers / TTS (existing)
|
||||
Ch 3+: Real callers (dynamically assigned per call)
|
||||
Ch N-1: Music (existing)
|
||||
Ch N: SFX (existing)
|
||||
```
|
||||
|
||||
### Call Flow — Real Caller
|
||||
|
||||
```
|
||||
Caller dials Twilio number
|
||||
→ Twilio POST /api/twilio/voice
|
||||
→ TwiML response: greeting + enqueue with hold music
|
||||
→ Caller waits in hold queue
|
||||
→ Host sees caller in dashboard queue panel
|
||||
→ Host clicks "Take Call"
|
||||
→ POST /api/queue/take/{call_sid}
|
||||
→ Twilio opens WebSocket to /api/twilio/stream
|
||||
→ Bidirectional audio:
|
||||
Caller audio → decode mulaw → dedicated Loopback channel
|
||||
Host audio + AI TTS → encode mulaw → Twilio → caller hears both
|
||||
→ Real-time Whisper transcription of caller audio
|
||||
→ Host hangs up → call summarized → stored in show history
|
||||
```
|
||||
|
||||
### Three-Way Call Flow
|
||||
|
||||
```
|
||||
Host mic ──────→ Ch 1 (recording)
|
||||
→ Twilio outbound (real caller hears you)
|
||||
→ Whisper transcription (AI gets your words)
|
||||
|
||||
Real caller ──→ Ch 3+ (recording, dedicated channel)
|
||||
→ Whisper transcription (AI gets their words)
|
||||
→ Host headphones
|
||||
|
||||
AI TTS ───────→ Ch 2 (recording)
|
||||
→ Twilio outbound (real caller hears AI)
|
||||
→ Host headphones (already works)
|
||||
```
|
||||
|
||||
Conversation history becomes three-party with role labels: `host`, `real_caller`, `ai_caller`.
|
||||
|
||||
### AI Auto-Respond Mode
|
||||
|
||||
When toggled on, after each real caller transcription chunk:
|
||||
|
||||
1. Lightweight LLM call ("should I respond?" — use fast model like Haiku)
|
||||
2. If YES → full response generated → TTS → plays on AI channel + streams to Twilio
|
||||
3. Cooldown (~10s) prevents rapid-fire
|
||||
4. Host can override with mute button
|
||||
|
||||
### AI Follow-Up System
|
||||
|
||||
After a real caller hangs up:
|
||||
|
||||
1. Full transcript (host + real caller + any AI) summarized by LLM
|
||||
2. Summary stored in `session.call_history`
|
||||
3. Next AI caller's system prompt includes show history:
|
||||
```
|
||||
EARLIER IN THE SHOW:
|
||||
- Dave (real caller) called about his wife leaving after 12 years.
|
||||
He got emotional about his kids.
|
||||
- Jasmine called about her boss hitting on her at work.
|
||||
You can reference these if it feels natural. Don't force it.
|
||||
```
|
||||
|
||||
**Host-triggered (default):** Click any AI caller as normal. They already have show context.
|
||||
|
||||
**Auto mode:** After real caller hangs up, system waits ~5-10s, picks a fitting AI caller via short LLM call, biases their background generation toward the topic, auto-connects.
|
||||
|
||||
## Backend Changes
|
||||
|
||||
### New Module: `backend/services/twilio_service.py`
|
||||
|
||||
Manages Twilio integration:
|
||||
- WebSocket handler for Media Streams (decode/encode mulaw 8kHz ↔ PCM)
|
||||
- Call queue state (waiting callers, SIDs, timestamps, assigned channels)
|
||||
- Channel pool management (allocate/release Loopback channels for real callers)
|
||||
- Outbound audio mixing (host + AI TTS → mulaw → Twilio)
|
||||
- Methods: `take_call()`, `hangup_real_caller()`, `get_queue()`, `send_audio_to_caller()`
|
||||
|
||||
### New Endpoints
|
||||
|
||||
```python
|
||||
# Twilio webhooks
|
||||
POST /api/twilio/voice # Incoming call → TwiML (greet + enqueue)
|
||||
POST /api/twilio/hold-music # Hold music TwiML for waiting callers
|
||||
WS /api/twilio/stream # Media Streams WebSocket (bidirectional audio)
|
||||
|
||||
# Host controls
|
||||
GET /api/queue # List waiting callers (number, wait time)
|
||||
POST /api/queue/take/{call_sid} # Dequeue caller → start media stream
|
||||
POST /api/queue/drop/{call_sid} # Drop caller from queue
|
||||
|
||||
# AI follow-up
|
||||
POST /api/followup/generate # Summarize last real call, trigger AI follow-up
|
||||
```
|
||||
|
||||
### Session Model Changes
|
||||
|
||||
```python
|
||||
class CallRecord:
|
||||
caller_type: str # "ai" or "real"
|
||||
caller_name: str # "Tony" or "Caller #3"
|
||||
summary: str # LLM-generated summary after hangup
|
||||
transcript: list[dict] # Full conversation [{role, content}]
|
||||
|
||||
class Session:
|
||||
# Existing fields...
|
||||
call_history: list[CallRecord] # All calls this episode
|
||||
active_real_caller: dict | None # {call_sid, phone, channel, name}
|
||||
active_ai_caller: str | None # Caller key
|
||||
ai_respond_mode: str # "manual" or "auto"
|
||||
auto_followup: bool # Auto-generate AI follow-up after real calls
|
||||
```
|
||||
|
||||
Three-party conversation history uses roles: `host`, `real_caller:{name}`, `ai_caller:{name}`.
|
||||
|
||||
### AI Caller Prompt Changes
|
||||
|
||||
`get_caller_prompt()` extended to include:
|
||||
- Show history from `session.call_history`
|
||||
- Current real caller context (if three-way call active)
|
||||
- Instructions for referencing real callers naturally
|
||||
|
||||
## Frontend Changes
|
||||
|
||||
### New: Call Queue Panel
|
||||
|
||||
Between callers section and chat. Shows waiting real callers with phone number and wait time. "Take Call" and "Drop" buttons per caller. Polls `/api/queue` every few seconds.
|
||||
|
||||
### Modified: Active Call Indicator
|
||||
|
||||
Shows real caller and AI caller simultaneously when both active:
|
||||
- Real caller: name, channel number, call duration, hang up button
|
||||
- AI caller: name, Manual/Auto toggle, "Let [name] respond" button (manual mode)
|
||||
- Auto Follow-Up checkbox
|
||||
|
||||
### Modified: Chat Log
|
||||
|
||||
Three-party with visual distinction:
|
||||
- Host messages: existing style
|
||||
- Real caller: labeled "Dave (caller)", distinct color
|
||||
- AI caller: labeled "Tony (AI)", distinct color
|
||||
|
||||
### Modified: Caller Grid
|
||||
|
||||
When real caller is active, clicking an AI caller adds them as third party instead of starting fresh call. Indicator shows which AI callers have been on the show this session.
|
||||
|
||||
## Dependencies
|
||||
|
||||
- `twilio` Python package (for TwiML generation, REST API)
|
||||
- Twilio account with phone number (~$1.15/mo + per-minute)
|
||||
- Cloudflare tunnel for exposing webhook endpoints
|
||||
- `audioop` or equivalent for mulaw encode/decode (stdlib in Python 3.11)
|
||||
|
||||
## Configuration
|
||||
|
||||
New env vars in `.env`:
|
||||
```
|
||||
TWILIO_ACCOUNT_SID=...
|
||||
TWILIO_AUTH_TOKEN=...
|
||||
TWILIO_PHONE_NUMBER=+1...
|
||||
TWILIO_WEBHOOK_BASE_URL=https://your-tunnel.cloudflare.com
|
||||
```
|
||||
1829
docs/plans/2026-02-05-real-callers-implementation.md
Normal file
1829
docs/plans/2026-02-05-real-callers-implementation.md
Normal file
File diff suppressed because it is too large
Load Diff
52
docs/plans/2026-02-05-signalwire-callers-design.md
Normal file
52
docs/plans/2026-02-05-signalwire-callers-design.md
Normal file
@@ -0,0 +1,52 @@
|
||||
# SignalWire Phone Call-In Design
|
||||
|
||||
## Goal
|
||||
|
||||
Replace browser-based WebSocket call-in with real phone calls via SignalWire. Callers dial 208-439-5853 and enter the show queue.
|
||||
|
||||
## Architecture
|
||||
|
||||
SignalWire handles PSTN connectivity. When a call comes in, SignalWire hits our webhook, we return XML telling it to open a bidirectional WebSocket stream with L16@16kHz audio. The audio flows through our existing pipeline — same queue, channel allocation, transcription, host mic streaming, and TTS streaming.
|
||||
|
||||
## Call Flow
|
||||
|
||||
1. Caller dials 208-439-5853
|
||||
2. SignalWire hits `POST /api/signalwire/voice` (via Cloudflare tunnel)
|
||||
3. We return `<Connect><Stream codec="L16@16000h">` XML
|
||||
4. SignalWire opens WebSocket to `/api/signalwire/stream`
|
||||
5. Caller enters queue — host sees phone number on dashboard
|
||||
6. Host takes call — audio flows bidirectionally
|
||||
7. Host hangs up — we call SignalWire REST API to end the phone call
|
||||
|
||||
## Audio Path
|
||||
|
||||
```
|
||||
Phone → PSTN → SignalWire → WebSocket (base64 L16 JSON) → Our server
|
||||
Our server → WebSocket (base64 L16 JSON) → SignalWire → PSTN → Phone
|
||||
```
|
||||
|
||||
## SignalWire WebSocket Protocol
|
||||
|
||||
Incoming: `{"event": "media", "media": {"payload": "<base64 L16 PCM 16kHz>"}}`
|
||||
Outgoing: `{"event": "media", "media": {"payload": "<base64 L16 PCM 16kHz>"}}`
|
||||
Start: `{"event": "start", "start": {"streamSid": "...", "callSid": "..."}}`
|
||||
Stop: `{"event": "stop"}`
|
||||
|
||||
## What Changes
|
||||
|
||||
- Remove: browser call-in page, browser WebSocket handler
|
||||
- Add: SignalWire webhook + WebSocket handler, hangup via REST API
|
||||
- Modify: CallerService (name→phone, base64 JSON encoding for send), dashboard (show phone number)
|
||||
- Unchanged: AudioService, queue logic, transcription, TTS streaming, three-way calls
|
||||
|
||||
## Config
|
||||
|
||||
```
|
||||
SIGNALWIRE_PROJECT_ID=8eb54732-ade3-4487-8b40-ecd2cd680df7
|
||||
SIGNALWIRE_SPACE=macneil-media-group-llc.signalwire.com
|
||||
SIGNALWIRE_TOKEN=PT9c9b61f44ee49914c614fed32aa5c3d7b9372b5199d81dec
|
||||
SIGNALWIRE_PHONE=+12084395853
|
||||
```
|
||||
|
||||
Webhook URL: `https://radioshow.macneilmediagroup.com/api/signalwire/voice`
|
||||
No SDK needed — httpx for the one REST call (hangup).
|
||||
855
docs/plans/2026-02-05-signalwire-implementation.md
Normal file
855
docs/plans/2026-02-05-signalwire-implementation.md
Normal file
@@ -0,0 +1,855 @@
|
||||
# SignalWire Phone Call-In Implementation Plan
|
||||
|
||||
> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.
|
||||
|
||||
**Goal:** Replace browser-based WebSocket call-in with real phone calls via SignalWire (208-439-5853).
|
||||
|
||||
**Architecture:** SignalWire hits our webhook on inbound calls, we return XML to open a bidirectional WebSocket stream with L16@16kHz audio. The existing queue, channel allocation, transcription, host mic streaming, and TTS streaming are reused — only the WebSocket message format changes (base64 JSON instead of raw binary).
|
||||
|
||||
**Tech Stack:** Python/FastAPI, SignalWire Compatibility API (LaML XML + WebSocket), httpx for REST calls, existing audio pipeline.
|
||||
|
||||
---
|
||||
|
||||
## Task 1: Add SignalWire Config
|
||||
|
||||
**Files:**
|
||||
- Modify: `backend/config.py`
|
||||
- Modify: `.env`
|
||||
|
||||
**Step 1: Add SignalWire settings to config.py**
|
||||
|
||||
In `backend/config.py`, add these fields to the `Settings` class after the existing API keys block (after line 16):
|
||||
|
||||
```python
|
||||
# SignalWire
|
||||
signalwire_project_id: str = os.getenv("SIGNALWIRE_PROJECT_ID", "")
|
||||
signalwire_space: str = os.getenv("SIGNALWIRE_SPACE", "")
|
||||
signalwire_token: str = os.getenv("SIGNALWIRE_TOKEN", "")
|
||||
signalwire_phone: str = os.getenv("SIGNALWIRE_PHONE", "")
|
||||
```
|
||||
|
||||
**Step 2: Add SignalWire vars to .env**
|
||||
|
||||
Append to `.env`:
|
||||
|
||||
```
|
||||
# SignalWire
|
||||
SIGNALWIRE_PROJECT_ID=8eb54732-ade3-4487-8b40-ecd2cd680df7
|
||||
SIGNALWIRE_SPACE=macneil-media-group-llc.signalwire.com
|
||||
SIGNALWIRE_TOKEN=PT9c9b61f44ee49914c614fed32aa5c3d7b9372b5199d81dec
|
||||
SIGNALWIRE_PHONE=+12084395853
|
||||
```
|
||||
|
||||
**Step 3: Verify config loads**
|
||||
|
||||
```bash
|
||||
cd /Users/lukemacneil/ai-podcast && python -c "from backend.config import settings; print(settings.signalwire_space)"
|
||||
```
|
||||
|
||||
Expected: `macneil-media-group-llc.signalwire.com`
|
||||
|
||||
**Step 4: Commit**
|
||||
|
||||
```bash
|
||||
git add backend/config.py .env
|
||||
git commit -m "Add SignalWire configuration"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 2: Update CallerService for SignalWire Protocol
|
||||
|
||||
**Files:**
|
||||
- Modify: `backend/services/caller_service.py`
|
||||
|
||||
The CallerService currently sends raw binary PCM frames. SignalWire needs base64-encoded L16 PCM wrapped in JSON. Also swap `name` field to `phone` since callers now have phone numbers.
|
||||
|
||||
**Step 1: Update queue to use `phone` instead of `name`**
|
||||
|
||||
In `caller_service.py`, make these changes:
|
||||
|
||||
1. Update docstring (line 1): `"""Phone caller queue and audio stream service"""`
|
||||
|
||||
2. In `add_to_queue` (line 24): Change parameter `name` to `phone`, and update the dict:
|
||||
|
||||
```python
|
||||
def add_to_queue(self, caller_id: str, phone: str):
|
||||
with self._lock:
|
||||
self._queue.append({
|
||||
"caller_id": caller_id,
|
||||
"phone": phone,
|
||||
"queued_at": time.time(),
|
||||
})
|
||||
print(f"[Caller] {phone} added to queue (ID: {caller_id})")
|
||||
```
|
||||
|
||||
3. In `get_queue` (line 38): Return `phone` instead of `name`:
|
||||
|
||||
```python
|
||||
def get_queue(self) -> list[dict]:
|
||||
now = time.time()
|
||||
with self._lock:
|
||||
return [
|
||||
{
|
||||
"caller_id": c["caller_id"],
|
||||
"phone": c["phone"],
|
||||
"wait_time": int(now - c["queued_at"]),
|
||||
}
|
||||
for c in self._queue
|
||||
]
|
||||
```
|
||||
|
||||
4. In `take_call` (line 62): Use `phone` instead of `name`:
|
||||
|
||||
```python
|
||||
def take_call(self, caller_id: str) -> dict:
|
||||
caller = None
|
||||
with self._lock:
|
||||
for c in self._queue:
|
||||
if c["caller_id"] == caller_id:
|
||||
caller = c
|
||||
break
|
||||
if caller:
|
||||
self._queue = [c for c in self._queue if c["caller_id"] != caller_id]
|
||||
|
||||
if not caller:
|
||||
raise ValueError(f"Caller {caller_id} not in queue")
|
||||
|
||||
channel = self.allocate_channel()
|
||||
self._caller_counter += 1
|
||||
phone = caller["phone"]
|
||||
|
||||
call_info = {
|
||||
"caller_id": caller_id,
|
||||
"phone": phone,
|
||||
"channel": channel,
|
||||
"started_at": time.time(),
|
||||
}
|
||||
self.active_calls[caller_id] = call_info
|
||||
print(f"[Caller] {phone} taken on air — channel {channel}")
|
||||
return call_info
|
||||
```
|
||||
|
||||
5. In `hangup` (line 89): Use `phone` instead of `name`:
|
||||
|
||||
```python
|
||||
def hangup(self, caller_id: str):
|
||||
call_info = self.active_calls.pop(caller_id, None)
|
||||
if call_info:
|
||||
self.release_channel(call_info["channel"])
|
||||
print(f"[Caller] {call_info['phone']} hung up — channel {call_info['channel']} released")
|
||||
self._websockets.pop(caller_id, None)
|
||||
```
|
||||
|
||||
**Step 2: Update `send_audio_to_caller` for SignalWire JSON format**
|
||||
|
||||
Replace the existing `send_audio_to_caller` method with:
|
||||
|
||||
```python
|
||||
async def send_audio_to_caller(self, caller_id: str, pcm_data: bytes, sample_rate: int):
|
||||
"""Send small audio chunk to caller via SignalWire WebSocket.
|
||||
Encodes L16 PCM as base64 JSON per SignalWire protocol.
|
||||
"""
|
||||
ws = self._websockets.get(caller_id)
|
||||
if not ws:
|
||||
return
|
||||
|
||||
try:
|
||||
import base64
|
||||
if sample_rate != 16000:
|
||||
audio = np.frombuffer(pcm_data, dtype=np.int16).astype(np.float32) / 32768.0
|
||||
ratio = 16000 / sample_rate
|
||||
out_len = int(len(audio) * ratio)
|
||||
indices = (np.arange(out_len) / ratio).astype(int)
|
||||
indices = np.clip(indices, 0, len(audio) - 1)
|
||||
audio = audio[indices]
|
||||
pcm_data = (audio * 32767).astype(np.int16).tobytes()
|
||||
|
||||
payload = base64.b64encode(pcm_data).decode('ascii')
|
||||
import json
|
||||
await ws.send_text(json.dumps({
|
||||
"event": "media",
|
||||
"media": {"payload": payload}
|
||||
}))
|
||||
except Exception as e:
|
||||
print(f"[Caller] Failed to send audio: {e}")
|
||||
```
|
||||
|
||||
**Step 3: Update `stream_audio_to_caller` for SignalWire JSON format**
|
||||
|
||||
Replace the existing `stream_audio_to_caller` method with:
|
||||
|
||||
```python
|
||||
async def stream_audio_to_caller(self, caller_id: str, pcm_data: bytes, sample_rate: int):
|
||||
"""Stream large audio (TTS) to caller in real-time chunks via SignalWire WebSocket."""
|
||||
ws = self._websockets.get(caller_id)
|
||||
if not ws:
|
||||
return
|
||||
|
||||
self.streaming_tts = True
|
||||
try:
|
||||
import base64
|
||||
import json
|
||||
audio = np.frombuffer(pcm_data, dtype=np.int16).astype(np.float32) / 32768.0
|
||||
if sample_rate != 16000:
|
||||
ratio = 16000 / sample_rate
|
||||
out_len = int(len(audio) * ratio)
|
||||
indices = (np.arange(out_len) / ratio).astype(int)
|
||||
indices = np.clip(indices, 0, len(audio) - 1)
|
||||
audio = audio[indices]
|
||||
|
||||
chunk_samples = 960
|
||||
for i in range(0, len(audio), chunk_samples):
|
||||
if caller_id not in self._websockets:
|
||||
break
|
||||
chunk = audio[i:i + chunk_samples]
|
||||
pcm_chunk = (chunk * 32767).astype(np.int16).tobytes()
|
||||
payload = base64.b64encode(pcm_chunk).decode('ascii')
|
||||
await ws.send_text(json.dumps({
|
||||
"event": "media",
|
||||
"media": {"payload": payload}
|
||||
}))
|
||||
await asyncio.sleep(0.055)
|
||||
|
||||
except Exception as e:
|
||||
print(f"[Caller] Failed to stream audio: {e}")
|
||||
finally:
|
||||
self.streaming_tts = False
|
||||
```
|
||||
|
||||
**Step 4: Remove `notify_caller` and `disconnect_caller` methods**
|
||||
|
||||
These sent browser-specific JSON control messages. SignalWire callers are disconnected via REST API (handled in main.py). Delete methods `notify_caller` (line 168) and `disconnect_caller` (line 175). They will be replaced with a REST-based hangup in Task 4.
|
||||
|
||||
**Step 5: Add `call_sid` tracking for SignalWire call hangup**
|
||||
|
||||
Add a dict to track SignalWire call SIDs so we can end calls via REST:
|
||||
|
||||
In `__init__`, after `self._websockets` line, add:
|
||||
|
||||
```python
|
||||
self._call_sids: dict[str, str] = {} # caller_id -> SignalWire callSid
|
||||
```
|
||||
|
||||
Add methods:
|
||||
|
||||
```python
|
||||
def register_call_sid(self, caller_id: str, call_sid: str):
|
||||
"""Track SignalWire callSid for a caller"""
|
||||
self._call_sids[caller_id] = call_sid
|
||||
|
||||
def get_call_sid(self, caller_id: str) -> str | None:
|
||||
"""Get SignalWire callSid for a caller"""
|
||||
return self._call_sids.get(caller_id)
|
||||
|
||||
def unregister_call_sid(self, caller_id: str):
|
||||
"""Remove callSid tracking"""
|
||||
self._call_sids.pop(caller_id, None)
|
||||
```
|
||||
|
||||
In `reset`, also clear `self._call_sids`:
|
||||
|
||||
```python
|
||||
self._call_sids.clear()
|
||||
```
|
||||
|
||||
In `hangup`, also clean up call_sid:
|
||||
|
||||
```python
|
||||
self._call_sids.pop(caller_id, None)
|
||||
```
|
||||
|
||||
**Step 6: Run existing tests**
|
||||
|
||||
```bash
|
||||
cd /Users/lukemacneil/ai-podcast && python -m pytest tests/test_caller_service.py -v
|
||||
```
|
||||
|
||||
Tests will likely need updates due to `name` → `phone` rename. Fix any failures.
|
||||
|
||||
**Step 7: Commit**
|
||||
|
||||
```bash
|
||||
git add backend/services/caller_service.py
|
||||
git commit -m "Update CallerService for SignalWire protocol"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 3: Add SignalWire Voice Webhook
|
||||
|
||||
**Files:**
|
||||
- Modify: `backend/main.py`
|
||||
|
||||
**Step 1: Add the voice webhook endpoint**
|
||||
|
||||
Add after the existing route definitions (after line 421), replacing the `/call-in` route:
|
||||
|
||||
```python
|
||||
# --- SignalWire Endpoints ---
|
||||
|
||||
from fastapi import Request, Response
|
||||
|
||||
@app.post("/api/signalwire/voice")
|
||||
async def signalwire_voice_webhook(request: Request):
|
||||
"""Handle inbound call from SignalWire — return XML to start bidirectional stream"""
|
||||
form = await request.form()
|
||||
caller_phone = form.get("From", "Unknown")
|
||||
call_sid = form.get("CallSid", "")
|
||||
print(f"[SignalWire] Inbound call from {caller_phone} (CallSid: {call_sid})")
|
||||
|
||||
# Build WebSocket URL from the request
|
||||
ws_scheme = "wss"
|
||||
host = request.headers.get("host", "radioshow.macneilmediagroup.com")
|
||||
stream_url = f"{ws_scheme}://{host}/api/signalwire/stream"
|
||||
|
||||
xml = f"""<?xml version="1.0" encoding="UTF-8"?>
|
||||
<Response>
|
||||
<Connect>
|
||||
<Stream url="{stream_url}" codec="L16@16000h">
|
||||
<Parameter name="caller_phone" value="{caller_phone}"/>
|
||||
<Parameter name="call_sid" value="{call_sid}"/>
|
||||
</Stream>
|
||||
</Connect>
|
||||
</Response>"""
|
||||
|
||||
return Response(content=xml, media_type="application/xml")
|
||||
```
|
||||
|
||||
**Step 2: Remove the `/call-in` route**
|
||||
|
||||
Delete these lines (around line 419-421):
|
||||
|
||||
```python
|
||||
@app.get("/call-in")
|
||||
async def call_in_page():
|
||||
return FileResponse(frontend_dir / "call-in.html")
|
||||
```
|
||||
|
||||
**Step 3: Verify server starts**
|
||||
|
||||
```bash
|
||||
cd /Users/lukemacneil/ai-podcast && python -c "from backend.main import app; print('OK')"
|
||||
```
|
||||
|
||||
**Step 4: Commit**
|
||||
|
||||
```bash
|
||||
git add backend/main.py
|
||||
git commit -m "Add SignalWire voice webhook, remove call-in route"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 4: Add SignalWire WebSocket Stream Handler
|
||||
|
||||
**Files:**
|
||||
- Modify: `backend/main.py`
|
||||
|
||||
This replaces the browser caller WebSocket handler at `/api/caller/stream`.
|
||||
|
||||
**Step 1: Replace the browser WebSocket handler**
|
||||
|
||||
Delete the entire `caller_audio_stream` function (the `@app.websocket("/api/caller/stream")` handler, lines 807-887).
|
||||
|
||||
Add the new SignalWire WebSocket handler:
|
||||
|
||||
```python
|
||||
@app.websocket("/api/signalwire/stream")
|
||||
async def signalwire_audio_stream(websocket: WebSocket):
|
||||
"""Handle SignalWire bidirectional audio stream"""
|
||||
await websocket.accept()
|
||||
|
||||
caller_id = str(uuid.uuid4())[:8]
|
||||
caller_phone = "Unknown"
|
||||
call_sid = ""
|
||||
audio_buffer = bytearray()
|
||||
CHUNK_DURATION_S = 3
|
||||
SAMPLE_RATE = 16000
|
||||
chunk_samples = CHUNK_DURATION_S * SAMPLE_RATE
|
||||
stream_started = False
|
||||
|
||||
try:
|
||||
while True:
|
||||
raw = await websocket.receive_text()
|
||||
msg = json.loads(raw)
|
||||
event = msg.get("event")
|
||||
|
||||
if event == "start":
|
||||
# Extract caller info from stream parameters
|
||||
params = {}
|
||||
for p in msg.get("start", {}).get("customParameters", {}):
|
||||
pass
|
||||
# customParameters comes as a dict
|
||||
custom = msg.get("start", {}).get("customParameters", {})
|
||||
caller_phone = custom.get("caller_phone", "Unknown")
|
||||
call_sid = custom.get("call_sid", "")
|
||||
|
||||
stream_started = True
|
||||
print(f"[SignalWire WS] Stream started: {caller_phone} (CallSid: {call_sid})")
|
||||
|
||||
# Add to queue and register
|
||||
caller_service.add_to_queue(caller_id, caller_phone)
|
||||
caller_service.register_websocket(caller_id, websocket)
|
||||
if call_sid:
|
||||
caller_service.register_call_sid(caller_id, call_sid)
|
||||
|
||||
elif event == "media" and stream_started:
|
||||
# Decode base64 L16 PCM audio
|
||||
import base64
|
||||
payload = msg.get("media", {}).get("payload", "")
|
||||
if not payload:
|
||||
continue
|
||||
|
||||
pcm_data = base64.b64decode(payload)
|
||||
|
||||
# Only process audio if caller is on air
|
||||
call_info = caller_service.active_calls.get(caller_id)
|
||||
if not call_info:
|
||||
continue
|
||||
|
||||
audio_buffer.extend(pcm_data)
|
||||
|
||||
# Route to configured live caller Loopback channel
|
||||
audio_service.route_real_caller_audio(pcm_data, SAMPLE_RATE)
|
||||
|
||||
# Transcribe when we have enough audio
|
||||
if len(audio_buffer) >= chunk_samples * 2:
|
||||
pcm_chunk = bytes(audio_buffer[:chunk_samples * 2])
|
||||
audio_buffer = audio_buffer[chunk_samples * 2:]
|
||||
asyncio.create_task(
|
||||
_handle_real_caller_transcription(caller_id, pcm_chunk, SAMPLE_RATE)
|
||||
)
|
||||
|
||||
elif event == "stop":
|
||||
print(f"[SignalWire WS] Stream stopped: {caller_phone}")
|
||||
break
|
||||
|
||||
except WebSocketDisconnect:
|
||||
print(f"[SignalWire WS] Disconnected: {caller_id} ({caller_phone})")
|
||||
except Exception as e:
|
||||
print(f"[SignalWire WS] Error: {e}")
|
||||
finally:
|
||||
caller_service.unregister_websocket(caller_id)
|
||||
caller_service.unregister_call_sid(caller_id)
|
||||
caller_service.remove_from_queue(caller_id)
|
||||
if caller_id in caller_service.active_calls:
|
||||
caller_service.hangup(caller_id)
|
||||
if session.active_real_caller and session.active_real_caller.get("caller_id") == caller_id:
|
||||
session.active_real_caller = None
|
||||
if len(caller_service.active_calls) == 0:
|
||||
audio_service.stop_host_stream()
|
||||
if audio_buffer:
|
||||
asyncio.create_task(
|
||||
_handle_real_caller_transcription(caller_id, bytes(audio_buffer), SAMPLE_RATE)
|
||||
)
|
||||
```
|
||||
|
||||
**Step 2: Commit**
|
||||
|
||||
```bash
|
||||
git add backend/main.py
|
||||
git commit -m "Add SignalWire WebSocket stream handler, remove browser handler"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 5: Update Hangup and Queue Endpoints for SignalWire
|
||||
|
||||
**Files:**
|
||||
- Modify: `backend/main.py`
|
||||
|
||||
When the host hangs up or drops a caller, we need to end the actual phone call via SignalWire's REST API.
|
||||
|
||||
**Step 1: Add SignalWire hangup helper**
|
||||
|
||||
Add this function near the top of `main.py` (after imports):
|
||||
|
||||
```python
|
||||
async def _signalwire_end_call(call_sid: str):
|
||||
"""End a phone call via SignalWire REST API"""
|
||||
if not call_sid or not settings.signalwire_space:
|
||||
return
|
||||
try:
|
||||
url = f"https://{settings.signalwire_space}/api/laml/2010-04-01/Accounts/{settings.signalwire_project_id}/Calls/{call_sid}"
|
||||
async with httpx.AsyncClient(timeout=10.0) as client:
|
||||
response = await client.post(
|
||||
url,
|
||||
data={"Status": "completed"},
|
||||
auth=(settings.signalwire_project_id, settings.signalwire_token),
|
||||
)
|
||||
print(f"[SignalWire] End call {call_sid}: {response.status_code}")
|
||||
except Exception as e:
|
||||
print(f"[SignalWire] Failed to end call {call_sid}: {e}")
|
||||
```
|
||||
|
||||
Also add `import httpx` at the top of main.py if not already present.
|
||||
|
||||
**Step 2: Update `take_call_from_queue`**
|
||||
|
||||
In the `take_call_from_queue` endpoint, update `name` references to `phone`:
|
||||
|
||||
```python
|
||||
@app.post("/api/queue/take/{caller_id}")
|
||||
async def take_call_from_queue(caller_id: str):
|
||||
"""Take a caller off hold and put them on air"""
|
||||
try:
|
||||
call_info = caller_service.take_call(caller_id)
|
||||
except ValueError as e:
|
||||
raise HTTPException(404, str(e))
|
||||
|
||||
session.active_real_caller = {
|
||||
"caller_id": call_info["caller_id"],
|
||||
"channel": call_info["channel"],
|
||||
"phone": call_info["phone"],
|
||||
}
|
||||
|
||||
# Start host mic streaming if this is the first real caller
|
||||
if len(caller_service.active_calls) == 1:
|
||||
_start_host_audio_sender()
|
||||
audio_service.start_host_stream(_host_audio_sync_callback)
|
||||
|
||||
return {
|
||||
"status": "on_air",
|
||||
"caller": call_info,
|
||||
}
|
||||
```
|
||||
|
||||
Note: The `notify_caller` call is removed — SignalWire callers don't need a JSON status message, they're already connected via the phone.
|
||||
|
||||
**Step 3: Update `drop_from_queue`**
|
||||
|
||||
End the phone call when dropping:
|
||||
|
||||
```python
|
||||
@app.post("/api/queue/drop/{caller_id}")
|
||||
async def drop_from_queue(caller_id: str):
|
||||
"""Drop a caller from the queue"""
|
||||
call_sid = caller_service.get_call_sid(caller_id)
|
||||
caller_service.remove_from_queue(caller_id)
|
||||
if call_sid:
|
||||
await _signalwire_end_call(call_sid)
|
||||
return {"status": "dropped"}
|
||||
```
|
||||
|
||||
**Step 4: Update `hangup_real_caller`**
|
||||
|
||||
End the phone call when hanging up:
|
||||
|
||||
```python
|
||||
@app.post("/api/hangup/real")
|
||||
async def hangup_real_caller():
|
||||
"""Hang up on real caller — disconnect immediately, summarize in background"""
|
||||
if not session.active_real_caller:
|
||||
raise HTTPException(400, "No active real caller")
|
||||
|
||||
caller_id = session.active_real_caller["caller_id"]
|
||||
caller_phone = session.active_real_caller["phone"]
|
||||
conversation_snapshot = list(session.conversation)
|
||||
auto_followup_enabled = session.auto_followup
|
||||
|
||||
# End the phone call via SignalWire
|
||||
call_sid = caller_service.get_call_sid(caller_id)
|
||||
caller_service.hangup(caller_id)
|
||||
if call_sid:
|
||||
asyncio.create_task(_signalwire_end_call(call_sid))
|
||||
|
||||
# Stop host streaming if no more active callers
|
||||
if len(caller_service.active_calls) == 0:
|
||||
audio_service.stop_host_stream()
|
||||
|
||||
session.active_real_caller = None
|
||||
|
||||
# Play hangup sound in background
|
||||
import threading
|
||||
hangup_sound = settings.sounds_dir / "hangup.wav"
|
||||
if hangup_sound.exists():
|
||||
threading.Thread(target=audio_service.play_sfx, args=(str(hangup_sound),), daemon=True).start()
|
||||
|
||||
# Summarize and store history in background
|
||||
asyncio.create_task(
|
||||
_summarize_real_call(caller_phone, conversation_snapshot, auto_followup_enabled)
|
||||
)
|
||||
|
||||
return {
|
||||
"status": "disconnected",
|
||||
"caller": caller_phone,
|
||||
}
|
||||
```
|
||||
|
||||
**Step 5: Update `_handle_real_caller_transcription`**
|
||||
|
||||
Change `caller_name` to `caller_phone`:
|
||||
|
||||
```python
|
||||
async def _handle_real_caller_transcription(caller_id: str, pcm_data: bytes, sample_rate: int):
|
||||
"""Transcribe a chunk of real caller audio and add to conversation"""
|
||||
call_info = caller_service.active_calls.get(caller_id)
|
||||
if not call_info:
|
||||
return
|
||||
|
||||
text = await transcribe_audio(pcm_data, source_sample_rate=sample_rate)
|
||||
if not text or not text.strip():
|
||||
return
|
||||
|
||||
caller_phone = call_info["phone"]
|
||||
print(f"[Real Caller] {caller_phone}: {text}")
|
||||
|
||||
session.add_message(f"real_caller:{caller_phone}", text)
|
||||
|
||||
if session.ai_respond_mode == "auto" and session.current_caller_key:
|
||||
asyncio.create_task(_check_ai_auto_respond(text, caller_phone))
|
||||
```
|
||||
|
||||
**Step 6: Update `_summarize_real_call`**
|
||||
|
||||
Change `caller_name` parameter to `caller_phone`:
|
||||
|
||||
```python
|
||||
async def _summarize_real_call(caller_phone: str, conversation: list, auto_followup_enabled: bool):
|
||||
"""Background task: summarize call and store in history"""
|
||||
summary = ""
|
||||
if conversation:
|
||||
transcript_text = "\n".join(
|
||||
f"{msg['role']}: {msg['content']}" for msg in conversation
|
||||
)
|
||||
summary = await llm_service.generate(
|
||||
messages=[{"role": "user", "content": f"Summarize this radio show call in 1-2 sentences:\n{transcript_text}"}],
|
||||
system_prompt="You summarize radio show conversations concisely. Focus on what the caller talked about and any emotional moments.",
|
||||
)
|
||||
|
||||
session.call_history.append(CallRecord(
|
||||
caller_type="real",
|
||||
caller_name=caller_phone,
|
||||
summary=summary,
|
||||
transcript=conversation,
|
||||
))
|
||||
print(f"[Real Caller] {caller_phone} call summarized: {summary[:80]}...")
|
||||
|
||||
if auto_followup_enabled:
|
||||
await _auto_followup(summary)
|
||||
```
|
||||
|
||||
**Step 7: Update `_check_ai_auto_respond`**
|
||||
|
||||
Change parameter name from `real_caller_name` to `real_caller_phone`:
|
||||
|
||||
```python
|
||||
async def _check_ai_auto_respond(real_caller_text: str, real_caller_phone: str):
|
||||
```
|
||||
|
||||
(The body doesn't use the name/phone parameter in any way that needs changing.)
|
||||
|
||||
**Step 8: Update TTS streaming references**
|
||||
|
||||
In `text_to_speech` endpoint and `_check_ai_auto_respond`, the `session.active_real_caller` dict now uses `phone` instead of `name`. No code change needed for the TTS streaming since it only uses `caller_id`.
|
||||
|
||||
**Step 9: Verify server starts**
|
||||
|
||||
```bash
|
||||
cd /Users/lukemacneil/ai-podcast && python -c "from backend.main import app; print('OK')"
|
||||
```
|
||||
|
||||
**Step 10: Commit**
|
||||
|
||||
```bash
|
||||
git add backend/main.py
|
||||
git commit -m "Update hangup and queue endpoints for SignalWire REST API"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 6: Update Frontend for Phone Callers
|
||||
|
||||
**Files:**
|
||||
- Modify: `frontend/js/app.js`
|
||||
- Modify: `frontend/index.html`
|
||||
|
||||
**Step 1: Update queue rendering in app.js**
|
||||
|
||||
In `renderQueue` function (around line 875), change `caller.name` to `caller.phone`:
|
||||
|
||||
```javascript
|
||||
el.innerHTML = queue.map(caller => {
|
||||
const mins = Math.floor(caller.wait_time / 60);
|
||||
const secs = caller.wait_time % 60;
|
||||
const waitStr = mins > 0 ? `${mins}m ${secs}s` : `${secs}s`;
|
||||
return `
|
||||
<div class="queue-item">
|
||||
<span class="queue-name">${caller.phone}</span>
|
||||
<span class="queue-wait">waiting ${waitStr}</span>
|
||||
<button class="queue-take-btn" onclick="takeCall('${caller.caller_id}')">Take Call</button>
|
||||
<button class="queue-drop-btn" onclick="dropCall('${caller.caller_id}')">Drop</button>
|
||||
</div>
|
||||
`;
|
||||
}).join('');
|
||||
```
|
||||
|
||||
**Step 2: Update `takeCall` log message**
|
||||
|
||||
In `takeCall` function (around line 896), change `data.caller.name` to `data.caller.phone`:
|
||||
|
||||
```javascript
|
||||
if (data.status === 'on_air') {
|
||||
showRealCaller(data.caller);
|
||||
log(`${data.caller.phone} is on air — Channel ${data.caller.channel}`);
|
||||
}
|
||||
```
|
||||
|
||||
**Step 3: Update `showRealCaller` to use phone**
|
||||
|
||||
In `showRealCaller` function (around line 939):
|
||||
|
||||
```javascript
|
||||
function showRealCaller(callerInfo) {
|
||||
const nameEl = document.getElementById('real-caller-name');
|
||||
const chEl = document.getElementById('real-caller-channel');
|
||||
if (nameEl) nameEl.textContent = callerInfo.phone;
|
||||
if (chEl) chEl.textContent = `Ch ${callerInfo.channel}`;
|
||||
```
|
||||
|
||||
**Step 4: Update index.html queue section header**
|
||||
|
||||
In `frontend/index.html`, change the queue section header (line 56) — remove the call-in page link:
|
||||
|
||||
```html
|
||||
<section class="queue-section">
|
||||
<h2>Incoming Calls</h2>
|
||||
<div id="call-queue" class="call-queue">
|
||||
```
|
||||
|
||||
**Step 5: Bump cache version in index.html**
|
||||
|
||||
Find the app.js script tag and bump the version:
|
||||
|
||||
```html
|
||||
<script src="/js/app.js?v=13"></script>
|
||||
```
|
||||
|
||||
**Step 6: Commit**
|
||||
|
||||
```bash
|
||||
git add frontend/js/app.js frontend/index.html
|
||||
git commit -m "Update frontend for phone caller display"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 7: Remove Browser Call-In Files
|
||||
|
||||
**Files:**
|
||||
- Delete: `frontend/call-in.html`
|
||||
- Delete: `frontend/js/call-in.js`
|
||||
|
||||
**Step 1: Delete files**
|
||||
|
||||
```bash
|
||||
cd /Users/lukemacneil/ai-podcast && rm frontend/call-in.html frontend/js/call-in.js
|
||||
```
|
||||
|
||||
**Step 2: Commit**
|
||||
|
||||
```bash
|
||||
git add frontend/call-in.html frontend/js/call-in.js
|
||||
git commit -m "Remove browser call-in page"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 8: Update Tests
|
||||
|
||||
**Files:**
|
||||
- Modify: `tests/test_caller_service.py`
|
||||
|
||||
**Step 1: Update tests for `name` → `phone` rename**
|
||||
|
||||
Throughout `test_caller_service.py`, change:
|
||||
- `add_to_queue(caller_id, "TestName")` → `add_to_queue(caller_id, "+15551234567")`
|
||||
- `caller["name"]` → `caller["phone"]`
|
||||
- `call_info["name"]` → `call_info["phone"]`
|
||||
|
||||
Also remove any tests for `notify_caller` or `disconnect_caller` if they exist, since those methods were removed.
|
||||
|
||||
**Step 2: Run all tests**
|
||||
|
||||
```bash
|
||||
cd /Users/lukemacneil/ai-podcast && python -m pytest tests/ -v
|
||||
```
|
||||
|
||||
Expected: All pass.
|
||||
|
||||
**Step 3: Commit**
|
||||
|
||||
```bash
|
||||
git add tests/
|
||||
git commit -m "Update tests for SignalWire phone caller format"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Task 9: Configure SignalWire Webhook and End-to-End Test
|
||||
|
||||
**Step 1: Start the server**
|
||||
|
||||
```bash
|
||||
cd /Users/lukemacneil/ai-podcast && python -m uvicorn backend.main:app --reload --host 0.0.0.0 --port 8000
|
||||
```
|
||||
|
||||
**Step 2: Verify webhook endpoint responds**
|
||||
|
||||
```bash
|
||||
curl -X POST http://localhost:8000/api/signalwire/voice \
|
||||
-d "From=+15551234567&CallSid=test123" \
|
||||
-H "Content-Type: application/x-www-form-urlencoded"
|
||||
```
|
||||
|
||||
Expected: XML response with `<Connect><Stream>` containing the WebSocket URL.
|
||||
|
||||
**Step 3: Verify Cloudflare tunnel is running**
|
||||
|
||||
```bash
|
||||
curl -s https://radioshow.macneilmediagroup.com/api/server/status
|
||||
```
|
||||
|
||||
Expected: JSON response with `"status": "running"`.
|
||||
|
||||
**Step 4: Configure SignalWire webhook**
|
||||
|
||||
In the SignalWire dashboard:
|
||||
1. Go to Phone Numbers → 208-439-5853
|
||||
2. Set "When a call comes in" to: `https://radioshow.macneilmediagroup.com/api/signalwire/voice`
|
||||
3. Method: POST
|
||||
4. Handler type: LaML Webhooks
|
||||
|
||||
**Step 5: Test with a real call**
|
||||
|
||||
Call 208-439-5853 from a phone. Expected:
|
||||
1. Call connects (no ringing/hold — goes straight to stream)
|
||||
2. Caller appears in queue on host dashboard with phone number
|
||||
3. Host clicks "Take Call" → audio flows bidirectionally
|
||||
4. Host clicks "Hang Up" → phone call ends
|
||||
|
||||
**Step 6: Commit any fixes needed**
|
||||
|
||||
```bash
|
||||
git add -A
|
||||
git commit -m "Final SignalWire integration fixes"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Summary
|
||||
|
||||
| Task | What | Key Files |
|
||||
|------|------|-----------|
|
||||
| 1 | SignalWire config | `config.py`, `.env` |
|
||||
| 2 | CallerService protocol update | `caller_service.py` |
|
||||
| 3 | Voice webhook endpoint | `main.py` |
|
||||
| 4 | WebSocket stream handler | `main.py` |
|
||||
| 5 | Hangup/queue via REST API | `main.py` |
|
||||
| 6 | Frontend phone display | `app.js`, `index.html` |
|
||||
| 7 | Remove browser call-in | `call-in.html`, `call-in.js` |
|
||||
| 8 | Update tests | `tests/` |
|
||||
| 9 | Configure & test | SignalWire dashboard |
|
||||
|
||||
Tasks 1-5 are sequential backend. Task 6-7 are frontend (can parallel after task 5). Task 8 after task 2. Task 9 is final integration test.
|
||||
@@ -593,6 +593,23 @@ section h2 {
|
||||
.hangup-btn.small { font-size: 0.75rem; padding: 0.2rem 0.5rem; }
|
||||
.auto-followup-label { display: flex; align-items: center; gap: 0.4rem; font-size: 0.8rem; color: #999; margin-top: 0.5rem; }
|
||||
|
||||
/* Returning Caller */
|
||||
.caller-btn.returning {
|
||||
border-color: #f9a825;
|
||||
color: #f9a825;
|
||||
}
|
||||
|
||||
.caller-btn.returning:hover {
|
||||
border-color: #fdd835;
|
||||
}
|
||||
|
||||
/* Screening Badges */
|
||||
.screening-badge { font-size: 0.7rem; padding: 0.1rem 0.4rem; border-radius: 3px; font-weight: bold; }
|
||||
.screening-badge.screening { background: #e65100; color: white; animation: pulse 1.5s infinite; }
|
||||
.screening-badge.screened { background: #2e7d32; color: white; }
|
||||
.screening-summary { font-size: 0.8rem; color: #aaa; font-style: italic; flex-basis: 100%; margin-top: 0.2rem; }
|
||||
.queue-item { flex-wrap: wrap; }
|
||||
|
||||
/* Three-Party Chat */
|
||||
.message.real-caller { border-left: 3px solid #c62828; padding-left: 0.5rem; }
|
||||
.message.ai-caller { border-left: 3px solid #1565c0; padding-left: 0.5rem; }
|
||||
|
||||
@@ -13,6 +13,7 @@
|
||||
<div class="header-buttons">
|
||||
<button id="on-air-btn" class="on-air-btn off">OFF AIR</button>
|
||||
<button id="new-session-btn" class="new-session-btn">New Session</button>
|
||||
<button id="export-session-btn">Export</button>
|
||||
<button id="settings-btn">Settings</button>
|
||||
</div>
|
||||
</header>
|
||||
|
||||
@@ -85,6 +85,9 @@ function initEventListeners() {
|
||||
});
|
||||
}
|
||||
|
||||
// Export session
|
||||
document.getElementById('export-session-btn')?.addEventListener('click', exportSession);
|
||||
|
||||
// Server controls
|
||||
document.getElementById('restart-server-btn')?.addEventListener('click', restartServer);
|
||||
document.getElementById('stop-server-btn')?.addEventListener('click', stopServer);
|
||||
@@ -351,7 +354,8 @@ async function loadCallers() {
|
||||
data.callers.forEach(caller => {
|
||||
const btn = document.createElement('button');
|
||||
btn.className = 'caller-btn';
|
||||
btn.textContent = caller.name;
|
||||
if (caller.returning) btn.classList.add('returning');
|
||||
btn.textContent = caller.returning ? `\u2605 ${caller.name}` : caller.name;
|
||||
btn.dataset.key = caller.key;
|
||||
btn.addEventListener('click', () => startCall(caller.key, caller.name));
|
||||
grid.appendChild(btn);
|
||||
@@ -996,10 +1000,21 @@ function renderQueue(queue) {
|
||||
const mins = Math.floor(caller.wait_time / 60);
|
||||
const secs = caller.wait_time % 60;
|
||||
const waitStr = mins > 0 ? `${mins}m ${secs}s` : `${secs}s`;
|
||||
const displayName = caller.caller_name || caller.phone;
|
||||
const screenBadge = caller.screening_status === 'complete'
|
||||
? '<span class="screening-badge screened">Screened</span>'
|
||||
: caller.screening_status === 'screening'
|
||||
? '<span class="screening-badge screening">Screening...</span>'
|
||||
: '';
|
||||
const summary = caller.screening_summary
|
||||
? `<div class="screening-summary">${caller.screening_summary}</div>`
|
||||
: '';
|
||||
return `
|
||||
<div class="queue-item">
|
||||
<span class="queue-name">${caller.phone}</span>
|
||||
<span class="queue-name">${displayName}</span>
|
||||
${screenBadge}
|
||||
<span class="queue-wait">waiting ${waitStr}</span>
|
||||
${summary}
|
||||
<button class="queue-take-btn" onclick="takeCall('${caller.caller_id}')">Take Call</button>
|
||||
<button class="queue-drop-btn" onclick="dropCall('${caller.caller_id}')">Drop</button>
|
||||
</div>
|
||||
@@ -1155,6 +1170,23 @@ async function fetchConversationUpdates() {
|
||||
}
|
||||
|
||||
|
||||
async function exportSession() {
|
||||
try {
|
||||
const res = await safeFetch('/api/session/export');
|
||||
const blob = new Blob([JSON.stringify(res, null, 2)], { type: 'application/json' });
|
||||
const url = URL.createObjectURL(blob);
|
||||
const a = document.createElement('a');
|
||||
a.href = url;
|
||||
a.download = `session-${res.session_id}.json`;
|
||||
a.click();
|
||||
URL.revokeObjectURL(url);
|
||||
log(`Exported session: ${res.call_count} calls`);
|
||||
} catch (err) {
|
||||
log('Export error: ' + err.message);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
async function stopServer() {
|
||||
if (!confirm('Stop the server? You will need to restart it manually.')) return;
|
||||
|
||||
|
||||
@@ -344,6 +344,7 @@ def main():
|
||||
parser.add_argument("--dry-run", "-d", action="store_true", help="Generate metadata but don't publish")
|
||||
parser.add_argument("--title", "-t", help="Override generated title")
|
||||
parser.add_argument("--description", help="Override generated description")
|
||||
parser.add_argument("--session-data", "-s", help="Path to session export JSON (from /api/session/export)")
|
||||
args = parser.parse_args()
|
||||
|
||||
audio_path = Path(args.audio_file).expanduser().resolve()
|
||||
@@ -358,12 +359,28 @@ def main():
|
||||
episode_number = get_next_episode_number()
|
||||
print(f"Episode number: {episode_number}")
|
||||
|
||||
# Load session data if provided
|
||||
session_data = None
|
||||
if args.session_data:
|
||||
session_path = Path(args.session_data).expanduser().resolve()
|
||||
if session_path.exists():
|
||||
with open(session_path) as f:
|
||||
session_data = json.load(f)
|
||||
print(f"Loaded session data: {session_data.get('call_count', 0)} calls")
|
||||
else:
|
||||
print(f"Warning: Session data file not found: {session_path}")
|
||||
|
||||
# Step 1: Transcribe
|
||||
transcript = transcribe_audio(str(audio_path))
|
||||
|
||||
# Step 2: Generate metadata
|
||||
metadata = generate_metadata(transcript, episode_number)
|
||||
|
||||
# Use session chapters if available (more accurate than LLM-generated)
|
||||
if session_data and session_data.get("chapters"):
|
||||
metadata["chapters"] = session_data["chapters"]
|
||||
print(f" Using {len(metadata['chapters'])} chapters from session data")
|
||||
|
||||
# Apply overrides
|
||||
if args.title:
|
||||
metadata["title"] = args.title
|
||||
@@ -374,6 +391,13 @@ def main():
|
||||
chapters_path = audio_path.with_suffix(".chapters.json")
|
||||
save_chapters(metadata, str(chapters_path))
|
||||
|
||||
# Save transcript alongside episode if session data available
|
||||
if session_data and session_data.get("transcript"):
|
||||
transcript_path = audio_path.with_suffix(".transcript.txt")
|
||||
with open(transcript_path, "w") as f:
|
||||
f.write(session_data["transcript"])
|
||||
print(f" Transcript saved to: {transcript_path}")
|
||||
|
||||
if args.dry_run:
|
||||
print("\n[DRY RUN] Would publish with:")
|
||||
print(f" Title: {metadata['title']}")
|
||||
|
||||
Reference in New Issue
Block a user