diff --git a/.gitignore b/.gitignore index f22713a..62c5b09 100644 --- a/.gitignore +++ b/.gitignore @@ -50,5 +50,9 @@ voices-v1.0.bin # Reference voices for TTS ref_audio/ +# YouTube OAuth credentials +youtube_client_secrets.json +youtube_token.json + # Claude settings (local) .claude/ diff --git a/make_clips.py b/make_clips.py index 71576c8..373e21b 100755 --- a/make_clips.py +++ b/make_clips.py @@ -31,8 +31,8 @@ load_dotenv(Path(__file__).parent / ".env") OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY") RSS_FEED_URL = "https://podcast.macneilmediagroup.com/@LukeAtTheRoost/feed.xml" EPISODE_CACHE_DIR = Path(__file__).parent / "clips" / ".episode-cache" -WHISPER_MODEL_FAST = "base" -WHISPER_MODEL_QUALITY = "large-v3" +WHISPER_MODEL_FAST = "distil-large-v3" +WHISPER_MODEL_QUALITY = "distil-large-v3" COVER_ART = Path(__file__).parent / "website" / "images" / "cover.png" # Fonts @@ -71,7 +71,7 @@ def _build_whisper_prompt(labeled_transcript: str) -> str: def transcribe_with_timestamps(audio_path: str, whisper_model: str = None, labeled_transcript: str = "") -> list[dict]: - """Transcribe audio with word-level timestamps using faster-whisper. + """Transcribe audio with word-level timestamps using mlx-whisper (Apple Silicon GPU). Returns list of segments: [{start, end, text, words: [{word, start, end}]}] """ @@ -83,43 +83,51 @@ def transcribe_with_timestamps(audio_path: str, whisper_model: str = None, return json.load(f) try: - from faster_whisper import WhisperModel + import mlx_whisper except ImportError: - print("Error: faster-whisper not installed. Run: pip install faster-whisper") + print("Error: mlx-whisper not installed. Run: pip install mlx-whisper") sys.exit(1) + MODEL_HF_REPOS = { + "distil-large-v3": "mlx-community/distil-whisper-large-v3", + "large-v3": "mlx-community/whisper-large-v3-mlx", + "medium": "mlx-community/whisper-medium-mlx", + "small": "mlx-community/whisper-small-mlx", + "base": "mlx-community/whisper-base-mlx", + } + hf_repo = MODEL_HF_REPOS.get(model_name, f"mlx-community/whisper-{model_name}-mlx") + initial_prompt = _build_whisper_prompt(labeled_transcript) - print(f" Model: {model_name}") + print(f" Model: {model_name} (MLX GPU)") if labeled_transcript: print(f" Prompt: {initial_prompt[:100]}...") - model = WhisperModel(model_name, compute_type="float32") - segments_iter, info = model.transcribe( + + result = mlx_whisper.transcribe( audio_path, + path_or_hf_repo=hf_repo, + language="en", word_timestamps=True, initial_prompt=initial_prompt, - language="en", - beam_size=5, - vad_filter=True, ) segments = [] - for seg in segments_iter: + for seg in result.get("segments", []): words = [] - if seg.words: - for w in seg.words: - words.append({ - "word": w.word.strip(), - "start": round(w.start, 3), - "end": round(w.end, 3), - }) + for w in seg.get("words", []): + words.append({ + "word": w["word"].strip(), + "start": round(w["start"], 3), + "end": round(w["end"], 3), + }) segments.append({ - "start": round(seg.start, 3), - "end": round(seg.end, 3), - "text": seg.text.strip(), + "start": round(seg["start"], 3), + "end": round(seg["end"], 3), + "text": seg["text"].strip(), "words": words, }) - print(f" Transcribed {info.duration:.1f}s ({len(segments)} segments)") + duration = segments[-1]["end"] if segments else 0 + print(f" Transcribed {duration:.1f}s ({len(segments)} segments)") with open(cache_path, "w") as f: json.dump(segments, f) @@ -131,33 +139,39 @@ def transcribe_with_timestamps(audio_path: str, whisper_model: str = None, def refine_clip_timestamps(audio_path: str, clips: list[dict], quality_model: str, labeled_transcript: str = "", ) -> dict[int, list[dict]]: - """Re-transcribe just the selected clip ranges with a high-quality model. + """Re-transcribe just the selected clip ranges with mlx-whisper (GPU). Extracts each clip segment, runs the quality model on it, and returns - refined segments with timestamps mapped back to the original timeline. + refined segments with word-level timestamps mapped back to the original timeline. Returns: {clip_index: [segments]} keyed by clip index """ try: - from faster_whisper import WhisperModel + import mlx_whisper except ImportError: - print("Error: faster-whisper not installed. Run: pip install faster-whisper") + print("Error: mlx-whisper not installed. Run: pip install mlx-whisper") sys.exit(1) - initial_prompt = _build_whisper_prompt(labeled_transcript) - print(f" Refinement model: {quality_model}") + MODEL_HF_REPOS = { + "distil-large-v3": "mlx-community/distil-whisper-large-v3", + "large-v3": "mlx-community/whisper-large-v3-mlx", + "medium": "mlx-community/whisper-medium-mlx", + "small": "mlx-community/whisper-small-mlx", + "base": "mlx-community/whisper-base-mlx", + } + hf_repo = MODEL_HF_REPOS.get(quality_model, f"mlx-community/whisper-{quality_model}-mlx") - model = None # Lazy-load so we skip if all cached + print(f" Refinement model: {quality_model} (MLX GPU)") + + initial_prompt = _build_whisper_prompt(labeled_transcript) refined = {} with tempfile.TemporaryDirectory() as tmp: for i, clip in enumerate(clips): - # Add padding around clip for context (Whisper does better with some lead-in) pad = 3.0 seg_start = max(0, clip["start_time"] - pad) seg_end = clip["end_time"] + pad - # Check cache first cache_key = f"{Path(audio_path).stem}_clip{i}_{seg_start:.1f}-{seg_end:.1f}" cache_path = Path(audio_path).parent / f".whisper_refine_{quality_model}_{cache_key}.json" if cache_path.exists(): @@ -166,7 +180,6 @@ def refine_clip_timestamps(audio_path: str, clips: list[dict], refined[i] = json.load(f) continue - # Extract clip segment to temp WAV seg_path = os.path.join(tmp, f"segment_{i}.wav") cmd = [ "ffmpeg", "-y", "-ss", str(seg_start), "-t", str(seg_end - seg_start), @@ -178,39 +191,35 @@ def refine_clip_timestamps(audio_path: str, clips: list[dict], refined[i] = [] continue - # Lazy-load model on first non-cached clip - if model is None: - model = WhisperModel(quality_model, compute_type="float32") - - segments_iter, info = model.transcribe( + mlx_result = mlx_whisper.transcribe( seg_path, + path_or_hf_repo=hf_repo, + language="en", word_timestamps=True, initial_prompt=initial_prompt, - language="en", - beam_size=5, - vad_filter=True, ) - # Collect segments and offset timestamps back to original timeline segments = [] - for seg in segments_iter: + for seg_data in mlx_result.get("segments", []): + text = seg_data["text"].strip() words = [] - if seg.words: - for w in seg.words: - words.append({ - "word": w.word.strip(), - "start": round(w.start + seg_start, 3), - "end": round(w.end + seg_start, 3), - }) + for w in seg_data.get("words", []): + words.append({ + "word": w["word"].strip(), + "start": round(w["start"] + seg_start, 3), + "end": round(w["end"] + seg_start, 3), + }) + segments.append({ - "start": round(seg.start + seg_start, 3), - "end": round(seg.end + seg_start, 3), - "text": seg.text.strip(), + "start": round(seg_data["start"] + seg_start, 3), + "end": round(seg_data["end"] + seg_start, 3), + "text": text, "words": words, }) refined[i] = segments - print(f" Clip {i+1}: Refined {info.duration:.1f}s → {len(segments)} segments") + seg_duration = segments[-1]["end"] - segments[0]["start"] if segments else 0 + print(f" Clip {i+1}: Refined {seg_duration:.1f}s → {len(segments)} segments") with open(cache_path, "w") as f: json.dump(segments, f) @@ -694,32 +703,116 @@ def _interpolate_speaker(idx: int, matched: dict, n_words: int) -> str | None: return None +def polish_clip_words(words: list[dict], labeled_transcript: str = "") -> list[dict]: + """Use LLM to fix punctuation, capitalization, and misheard words. + + Sends the raw whisper words to an LLM, gets back a corrected version, + and maps corrections back to the original timed words. + """ + if not words or not OPENROUTER_API_KEY: + return words + + raw_text = " ".join(w["word"] for w in words) + + context = "" + if labeled_transcript: + context = f"\nFor reference, here's the speaker-labeled transcript of this section (use it to correct misheard words and names):\n{labeled_transcript[:3000]}\n" + + prompt = f"""Fix this podcast transcript excerpt so it reads as proper sentences. Fix punctuation, capitalization, and obvious misheard words. + +RULES: +- Keep the EXACT same number of words in the EXACT same order +- Only change capitalization, punctuation attached to words, and obvious mishearings +- Do NOT add, remove, merge, or reorder words +- Contractions count as one word (don't = 1 word) +- Return ONLY the corrected text, nothing else +{context} +RAW TEXT ({len(words)} words): +{raw_text}""" + + try: + response = requests.post( + "https://openrouter.ai/api/v1/chat/completions", + headers={ + "Authorization": f"Bearer {OPENROUTER_API_KEY}", + "Content-Type": "application/json", + }, + json={ + "model": "anthropic/claude-sonnet-4-5", + "messages": [{"role": "user", "content": prompt}], + "max_tokens": 2048, + "temperature": 0, + }, + timeout=30, + ) + if response.status_code != 200: + print(f" Polish failed ({response.status_code}), using raw text") + return words + + polished = response.json()["choices"][0]["message"]["content"].strip() + polished_words = polished.split() + + if len(polished_words) != len(words): + print(f" Polish word count mismatch ({len(polished_words)} vs {len(words)}), using raw text") + return words + + changes = 0 + for i, pw in enumerate(polished_words): + if pw != words[i]["word"]: + changes += 1 + words[i]["word"] = pw + + if changes: + print(f" Polished {changes} words") + + except Exception as e: + print(f" Polish error: {e}") + + return words + + def group_words_into_lines(words: list[dict], clip_start: float, clip_duration: float) -> list[dict]: """Group words into timed caption lines for rendering. + Splits at speaker changes so each line has a single, correct speaker label. Returns list of: {start, end, speaker, words: [{word, highlighted}]} """ if not words: return [] - # Group words into display lines (5-7 words per line) - raw_lines = [] - current_line = [] + # First split at speaker boundaries, then group into display lines + speaker_groups = [] + current_group = [] + current_speaker = words[0].get("speaker", "") for w in words: - current_line.append(w) - if len(current_line) >= 6 or w["word"].rstrip().endswith(('.', '?', '!', ',')): - if len(current_line) >= 3: - raw_lines.append(current_line) - current_line = [] - if current_line: - if raw_lines and len(current_line) < 3: - raw_lines[-1].extend(current_line) - else: - raw_lines.append(current_line) + speaker = w.get("speaker", "") + if speaker and speaker != current_speaker and current_group: + speaker_groups.append((current_speaker, current_group)) + current_group = [] + current_speaker = speaker + current_group.append(w) + if current_group: + speaker_groups.append((current_speaker, current_group)) + + # Now group each speaker's words into display lines (5-7 words) + raw_lines = [] + for speaker, group_words in speaker_groups: + current_line = [] + for w in group_words: + current_line.append(w) + if len(current_line) >= 6 or w["word"].rstrip().endswith(('.', '?', '!', ',')): + if len(current_line) >= 3: + raw_lines.append((speaker, current_line)) + current_line = [] + if current_line: + if raw_lines and len(current_line) < 3 and raw_lines[-1][0] == speaker: + raw_lines[-1] = (speaker, raw_lines[-1][1] + current_line) + else: + raw_lines.append((speaker, current_line)) lines = [] - for line_words in raw_lines: + for speaker, line_words in raw_lines: line_start = line_words[0]["start"] - clip_start line_end = line_words[-1]["end"] - clip_start @@ -733,7 +826,7 @@ def group_words_into_lines(words: list[dict], clip_start: float, lines.append({ "start": line_start, "end": line_end, - "speaker": line_words[0].get("speaker", ""), + "speaker": speaker, "words": line_words, }) @@ -1334,6 +1427,9 @@ def main(): clip["start_time"], clip["end_time"], word_source) + # Polish text with LLM (fix punctuation, capitalization, mishearings) + clip_words = polish_clip_words(clip_words, labeled_transcript) + # Group words into timed caption lines caption_lines = group_words_into_lines( clip_words, clip["start_time"], duration diff --git a/upload_clips.py b/upload_clips.py old mode 100644 new mode 100755 index 6c34a6b..a08875c --- a/upload_clips.py +++ b/upload_clips.py @@ -1,12 +1,11 @@ #!/usr/bin/env python3 -"""Upload podcast clips to social media via Postiz (and direct Bluesky via atproto). +"""Upload podcast clips to social media (direct YouTube & Bluesky, Postiz for others). Usage: - python upload_clips.py clips/episode-12/ - python upload_clips.py clips/episode-12/ --clip 1 - python upload_clips.py clips/episode-12/ --platforms ig,yt - python upload_clips.py clips/episode-12/ --schedule "2026-02-16T10:00:00" - python upload_clips.py clips/episode-12/ --yes # skip confirmation + python upload_clips.py # interactive: pick episode, clips, platforms + python upload_clips.py clips/episode-12/ # pick clips and platforms interactively + python upload_clips.py clips/episode-12/ --clip 1 --platforms ig,yt + python upload_clips.py clips/episode-12/ --yes # skip all prompts, upload everything """ import argparse @@ -27,6 +26,9 @@ POSTIZ_URL = os.getenv("POSTIZ_URL", "https://social.lukeattheroost.com") BSKY_HANDLE = os.getenv("BSKY_HANDLE", "lukeattheroost.bsky.social") BSKY_APP_PASSWORD = os.getenv("BSKY_APP_PASSWORD") +YT_CLIENT_SECRETS = Path(__file__).parent / "youtube_client_secrets.json" +YT_TOKEN_FILE = Path(__file__).parent / "youtube_token.json" + PLATFORM_ALIASES = { "ig": "instagram", "insta": "instagram", "instagram": "instagram", "yt": "youtube", "youtube": "youtube", @@ -214,6 +216,106 @@ def post_to_bluesky(clip: dict, clip_file: Path) -> bool: return True +def get_youtube_service(): + """Authenticate with YouTube API. First run opens a browser, then reuses saved token.""" + from google.oauth2.credentials import Credentials + from google_auth_oauthlib.flow import InstalledAppFlow + from google.auth.transport.requests import Request + from googleapiclient.discovery import build as yt_build + + scopes = ["https://www.googleapis.com/auth/youtube.upload"] + creds = None + + if YT_TOKEN_FILE.exists(): + creds = Credentials.from_authorized_user_file(str(YT_TOKEN_FILE), scopes) + + if not creds or not creds.valid: + if creds and creds.expired and creds.refresh_token: + creds.refresh(Request()) + else: + if not YT_CLIENT_SECRETS.exists(): + print(" Error: youtube_client_secrets.json not found") + print(" Download OAuth2 Desktop App credentials from Google Cloud Console") + return None + flow = InstalledAppFlow.from_client_secrets_file(str(YT_CLIENT_SECRETS), scopes) + creds = flow.run_local_server(port=8090) + + with open(YT_TOKEN_FILE, "w") as f: + f.write(creds.to_json()) + + return yt_build("youtube", "v3", credentials=creds) + + +def post_to_youtube(clip: dict, clip_file: Path) -> bool: + """Upload a clip directly to YouTube Shorts via the Data API.""" + import time + import random + from googleapiclient.http import MediaFileUpload + from googleapiclient.errors import HttpError + + youtube = get_youtube_service() + if not youtube: + return False + + title = clip["title"] + if "#Shorts" not in title: + title = f"{title} #Shorts" + + description = build_content(clip, "youtube") + if "#Shorts" not in description: + description += "\n\n#Shorts" + + tags = [h.lstrip("#") for h in clip.get("hashtags", [])] + if "Shorts" not in tags: + tags.insert(0, "Shorts") + + body = { + "snippet": { + "title": title[:100], + "description": description, + "tags": tags, + "categoryId": "24", # Entertainment + }, + "status": { + "privacyStatus": "public", + "selfDeclaredMadeForKids": False, + }, + } + + media = MediaFileUpload( + str(clip_file), + mimetype="video/mp4", + chunksize=256 * 1024, + resumable=True, + ) + + request = youtube.videos().insert(part="snippet,status", body=body, media_body=media) + + file_size = clip_file.stat().st_size / 1_000_000 + print(f" Uploading video ({file_size:.1f} MB)...") + + response = None + retry = 0 + while response is None: + try: + status, response = request.next_chunk() + if status: + print(f" Upload {int(status.progress() * 100)}%...") + except HttpError as e: + if e.resp.status in (500, 502, 503, 504) and retry < 5: + retry += 1 + wait = random.random() * (2 ** retry) + print(f" Retrying in {wait:.1f}s...") + time.sleep(wait) + else: + print(f" YouTube API error: {e}") + return False + + video_id = response["id"] + print(f" https://youtube.com/shorts/{video_id}") + return True + + def create_post(integration_id: str, content: str, media: dict, settings: dict, schedule: str | None = None) -> dict: from datetime import datetime, timezone @@ -253,7 +355,7 @@ def create_post(integration_id: str, content: str, media: dict, def main(): valid_names = sorted(set(PLATFORM_ALIASES.keys())) parser = argparse.ArgumentParser(description="Upload podcast clips to social media via Postiz") - parser.add_argument("clips_dir", help="Path to clips directory (e.g. clips/episode-12/)") + parser.add_argument("clips_dir", nargs="?", help="Path to clips directory (e.g. clips/episode-12/). If omitted, shows a picker.") parser.add_argument("--clip", "-c", type=int, help="Upload only clip N (1-indexed)") parser.add_argument("--platforms", "-p", help=f"Comma-separated platforms ({','.join(ALL_PLATFORMS)}). Default: all") @@ -266,6 +368,75 @@ def main(): print("Error: POSTIZ_API_KEY not set in .env") sys.exit(1) + # Resolve clips directory — pick interactively if not provided + if args.clips_dir: + clips_dir = Path(args.clips_dir).expanduser().resolve() + else: + clips_root = Path(__file__).parent / "clips" + episode_dirs = sorted( + [d for d in clips_root.iterdir() + if d.is_dir() and not d.name.startswith(".") and (d / "clips-metadata.json").exists()], + key=lambda d: d.name, + ) + if not episode_dirs: + print("No clip directories found in clips/. Run make_clips.py first.") + sys.exit(1) + print("\nAvailable episodes:\n") + for i, d in enumerate(episode_dirs): + with open(d / "clips-metadata.json") as f: + meta = json.load(f) + print(f" {i+1}. {d.name} ({len(meta)} clip{'s' if len(meta) != 1 else ''})") + print() + while True: + try: + choice = input("Which episode? ").strip() + idx = int(choice) - 1 + if 0 <= idx < len(episode_dirs): + clips_dir = episode_dirs[idx] + break + print(f" Enter 1-{len(episode_dirs)}") + except (ValueError, EOFError): + print(f" Enter an episode number") + + metadata_path = clips_dir / "clips-metadata.json" + if not metadata_path.exists(): + print(f"Error: No clips-metadata.json found in {clips_dir}") + print("Run make_clips.py first to generate clips and metadata.") + sys.exit(1) + + with open(metadata_path) as f: + clips = json.load(f) + + # Pick clips + if args.clip: + if args.clip < 1 or args.clip > len(clips): + print(f"Error: Clip {args.clip} not found (have {len(clips)} clips)") + sys.exit(1) + clips = [clips[args.clip - 1]] + elif not args.yes: + print(f"\nFound {len(clips)} clip(s):\n") + for i, clip in enumerate(clips): + desc = clip.get('description', clip.get('caption_text', '')) + if len(desc) > 70: + desc = desc[:desc.rfind(' ', 0, 70)] + '...' + print(f" {i+1}. \"{clip['title']}\" ({clip['duration']:.0f}s)") + print(f" {desc}") + print(f"\n a. All clips") + print() + while True: + choice = input("Which clips? (e.g. 1,3 or a for all): ").strip().lower() + if choice in ('a', 'all'): + break + try: + indices = [int(x.strip()) for x in choice.split(",")] + if all(1 <= x <= len(clips) for x in indices): + clips = [clips[x - 1] for x in indices] + break + print(f" Invalid selection. Enter 1-{len(clips)}, comma-separated, or 'a' for all.") + except (ValueError, EOFError): + print(f" Enter clip numbers (e.g. 1,3) or 'a' for all") + + # Pick platforms if args.platforms: requested = [] for p in args.platforms.split(","): @@ -276,28 +447,29 @@ def main(): sys.exit(1) requested.append(PLATFORM_ALIASES[p]) target_platforms = list(dict.fromkeys(requested)) + elif not args.yes: + print(f"\nPlatforms:\n") + for i, p in enumerate(ALL_PLATFORMS): + print(f" {i+1}. {PLATFORM_DISPLAY[p]}") + print(f"\n a. All platforms (default)") + print() + choice = input("Which platforms? (e.g. 1,3,5 or a for all) [a]: ").strip().lower() + if choice and choice not in ('a', 'all'): + try: + indices = [int(x.strip()) for x in choice.split(",")] + target_platforms = [ALL_PLATFORMS[x - 1] for x in indices if 1 <= x <= len(ALL_PLATFORMS)] + if not target_platforms: + target_platforms = ALL_PLATFORMS[:] + except (ValueError, IndexError): + target_platforms = ALL_PLATFORMS[:] + else: + target_platforms = ALL_PLATFORMS[:] else: target_platforms = ALL_PLATFORMS[:] - clips_dir = Path(args.clips_dir).expanduser().resolve() - metadata_path = clips_dir / "clips-metadata.json" - - if not metadata_path.exists(): - print(f"Error: No clips-metadata.json found in {clips_dir}") - print("Run make_clips.py first to generate clips and metadata.") - sys.exit(1) - - with open(metadata_path) as f: - clips = json.load(f) - - if args.clip: - if args.clip < 1 or args.clip > len(clips): - print(f"Error: Clip {args.clip} not found (have {len(clips)} clips)") - sys.exit(1) - clips = [clips[args.clip - 1]] - + DIRECT_PLATFORMS = {"bluesky", "youtube"} needs_postiz = not args.dry_run and any( - p != "bluesky" for p in target_platforms) + p not in DIRECT_PLATFORMS for p in target_platforms) if needs_postiz: print("Fetching connected accounts from Postiz...") integrations = fetch_integrations() @@ -312,6 +484,12 @@ def main(): else: print("Warning: BSKY_APP_PASSWORD not set in .env, skipping Bluesky") continue + if platform == "youtube": + if YT_CLIENT_SECRETS.exists() or YT_TOKEN_FILE.exists() or args.dry_run: + active_platforms[platform] = {"name": "YouTube Shorts", "_direct": True} + else: + print("Warning: youtube_client_secrets.json not found, skipping YouTube") + continue if args.dry_run: active_platforms[platform] = {"name": PLATFORM_DISPLAY[platform]} continue @@ -384,6 +562,16 @@ def main(): else: print(f" {display}: Failed") + if "youtube" in active_platforms: + print(f" Posting to YouTube Shorts (direct)...") + try: + if post_to_youtube(clip, clip_file): + print(f" YouTube: Posted!") + else: + print(f" YouTube: Failed") + except Exception as e: + print(f" YouTube: Failed — {e}") + if "bluesky" in active_platforms: print(f" Posting to Bluesky (direct)...") try: diff --git a/website/css/style.css b/website/css/style.css index 639fe27..9621d7f 100644 --- a/website/css/style.css +++ b/website/css/style.css @@ -47,7 +47,7 @@ a:hover { /* Hero */ .hero { - padding: 3rem 1.5rem 2rem; + padding: 3rem 1.5rem 2.5rem; max-width: 900px; margin: 0 auto; text-align: center; @@ -57,14 +57,14 @@ a:hover { display: flex; flex-direction: column; align-items: center; - gap: 2rem; + gap: 1.5rem; } .cover-art { - width: 220px; - height: 220px; + width: 260px; + height: 260px; border-radius: var(--radius); - box-shadow: 0 8px 32px rgba(232, 121, 29, 0.35); + box-shadow: 0 8px 32px rgba(232, 121, 29, 0.25); object-fit: cover; } @@ -72,31 +72,32 @@ a:hover { display: flex; flex-direction: column; align-items: center; - gap: 0.75rem; + gap: 0.5rem; } .hero h1 { - font-size: 2.5rem; + font-size: 2.8rem; font-weight: 800; letter-spacing: -0.02em; } .tagline { - font-size: 1.15rem; + font-size: 1.2rem; color: var(--text-muted); - max-width: 400px; + max-width: 500px; + line-height: 1.5; } .phone { display: flex; align-items: center; justify-content: center; - gap: 0.6rem; - margin-top: 0.5rem; + gap: 0.5rem; + margin-top: 0.25rem; } .phone-inline { - font-size: 0.95rem; + font-size: 1rem; color: var(--text-muted); } @@ -110,12 +111,12 @@ a:hover { .on-air-badge { display: none; align-items: center; - gap: 0.4rem; + gap: 0.35rem; background: var(--accent-red); color: #fff; - padding: 0.25rem 0.75rem; + padding: 0.2rem 0.6rem; border-radius: 50px; - font-size: 0.7rem; + font-size: 0.65rem; font-weight: 800; letter-spacing: 0.12em; text-transform: uppercase; @@ -128,8 +129,8 @@ a:hover { } .on-air-dot { - width: 8px; - height: 8px; + width: 7px; + height: 7px; border-radius: 50%; background: #fff; animation: on-air-blink 1s step-end infinite; @@ -149,11 +150,11 @@ a:hover { .off-air-badge { display: inline-flex; align-items: center; - background: #444; + background: rgba(255, 255, 255, 0.08); color: var(--text-muted); - padding: 0.25rem 0.75rem; + padding: 0.2rem 0.6rem; border-radius: 50px; - font-size: 0.7rem; + font-size: 0.65rem; font-weight: 700; letter-spacing: 0.1em; text-transform: uppercase; @@ -169,52 +170,58 @@ a:hover { text-shadow: 0 0 16px rgba(204, 34, 34, 0.35); } -/* Subscribe buttons — primary listen platforms */ +/* Subscribe — compact inline text links */ .subscribe-row { display: flex; - flex-direction: column; align-items: center; - gap: 0.6rem; - margin-top: 1.5rem; + justify-content: center; + gap: 0.5rem; + margin-top: 1rem; } .subscribe-label { - font-size: 0.75rem; + font-size: 0.8rem; color: var(--text-muted); text-transform: uppercase; - letter-spacing: 0.15em; + letter-spacing: 0.12em; + opacity: 0.6; } .subscribe-buttons { display: flex; flex-wrap: wrap; justify-content: center; - gap: 0.5rem; + align-items: center; + gap: 0.15rem; } .subscribe-btn { display: inline-flex; align-items: center; - gap: 0.4rem; - padding: 0.45rem 1rem; - border-radius: 50px; - font-size: 0.8rem; + gap: 0.35rem; + padding: 0.4rem 0.75rem; + border-radius: 6px; + font-size: 0.9rem; font-weight: 600; - color: var(--text); + color: var(--text-muted); background: transparent; - border: 1px solid var(--text-muted); - transition: border-color 0.2s, color 0.2s; + border: none; + transition: color 0.2s; } .subscribe-btn:hover { - border-color: var(--accent); color: var(--accent); } .subscribe-btn svg { - width: 14px; - height: 14px; + width: 15px; + height: 15px; flex-shrink: 0; + opacity: 0.6; +} + +.subscribe-btn:hover svg { + opacity: 1; } /* Secondary links — How It Works, Discord, Support */ @@ -224,23 +231,35 @@ a:hover { justify-content: center; align-items: center; gap: 0.5rem; - margin-top: 0.75rem; + margin-top: 0.25rem; } .secondary-link { - font-size: 0.8rem; + font-size: 0.85rem; color: var(--text-muted); - transition: color 0.2s; + opacity: 0.6; + transition: color 0.2s, opacity 0.2s; } .secondary-link:hover { color: var(--accent); + opacity: 1; } .secondary-sep { color: var(--text-muted); - opacity: 0.4; - font-size: 0.8rem; + opacity: 0.3; + font-size: 0.85rem; +} + +.support-link { + color: var(--accent); + opacity: 1; + font-weight: 600; +} + +.support-link:hover { + color: var(--accent-hover); } /* Episodes */ @@ -1201,13 +1220,15 @@ a:hover { /* Desktop */ @media (min-width: 768px) { .hero { - padding: 4rem 2rem 2.5rem; + padding: 3.5rem 2rem 2.5rem; + max-width: 1000px; } .hero-inner { flex-direction: row; text-align: left; - gap: 3rem; + gap: 2.5rem; + align-items: center; } .hero-info { @@ -1215,8 +1236,17 @@ a:hover { } .cover-art { - width: 260px; - height: 260px; + width: 280px; + height: 280px; + flex-shrink: 0; + } + + .hero h1 { + font-size: 2.8rem; + } + + .phone { + justify-content: flex-start; } .subscribe-row { diff --git a/website/episode.html b/website/episode.html index 441ee7b..a8c33ba 100644 --- a/website/episode.html +++ b/website/episode.html @@ -109,9 +109,9 @@ YouTube -

Support the show: Ko-fi

+

Support the Show

Sales & Collaboration: luke@macneilmediagroup.com

-

© 2026 Luke at the Roost · Privacy Policy

+

© 2026 Luke at the Roost · Privacy Policy · System Status

diff --git a/website/how-it-works.html b/website/how-it-works.html index 20ddf90..d1dc78e 100644 --- a/website/how-it-works.html +++ b/website/how-it-works.html @@ -69,6 +69,109 @@

Every caller on the show is a one-of-a-kind character — generated in real time by a custom-built AI system. Here's a peek behind the curtain.

+ +
+

The Anatomy of an AI Caller

+ +
+
+
1
+
+

A Person Is Born

+

Every caller starts as a blank slate. The system generates a complete identity: name, age, job, hometown, and personality. Each caller gets a unique speaking style — some ramble, some are blunt, some deflect with humor. They have relationships, vehicles, strong food opinions, nostalgic memories, and reasons for being up this late. They know what they were watching on TV, what errand they ran today, and what song was on the radio before they called.

+

Some callers become regulars. The system tracks returning callers across episodes — they remember past conversations, reference things they talked about before, and their stories evolve over time. You'll hear Carla update you on her divorce, or Carl check in about his gambling recovery. They're not reset between shows.

+

And some callers are drunk, high, or flat-out unhinged. They'll call with conspiracy theories about pigeons being government drones, existential crises about whether fish know they're wet, or to confess they accidentally set their kitchen on fire trying to make grilled cheese at 3 AM.

+
+
+ Unique Names + 320 +
+
+ Personality Layers + 189+ +
+
+ Towns with Real Knowledge + 20 +
+
+ Returning Regulars + 12 callers +
+
+
+
+ +
+
2
+
+

They Know Their World

+

Callers know real facts about where they live — the restaurants, the highways, the local gossip. When a caller says they're from Lordsburg, they actually know about the Shakespeare ghost town and the drive to Deming. They know the current weather outside their window, what day of the week it is, whether it's monsoon season or chile harvest. They have strong opinions about where to get the best green chile and get nostalgic about how their town used to be. The system also pulls in real-time news so callers can reference things that actually happened today.

+
+
+ +
+
3
+
+

They Have a Reason to Call

+

Some callers have a problem — a fight with a neighbor, a situation at work, something weighing on them at 2 AM. Others call to geek out about Severance, argue about poker strategy, or share something they read about quantum physics. The system draws from over 570 discussion topics across dozens of categories and more than 1,400 life scenarios. Every caller has a purpose, not just a script.

+
+
+ 70% + Need advice +
+
+ 30% + Want to talk about something +
+
+
+
+ +
+
4
+
+

The Conversation Is Real

+

Luke talks to each caller using push-to-talk, just like a real radio show. His voice is transcribed in real time, sent to an AI that responds in character, and then converted to speech using a voice engine — all in a few seconds. The AI doesn't just answer questions; it reacts, gets emotional, goes on tangents, and remembers what was said earlier in the show. Callers even react to previous callers — "Hey Luke, I heard that guy Tony earlier and I got to say, he's full of it." It makes the show feel like a living community, not isolated calls.

+
+
+ +
+
5
+
+

Real Callers Call In Too

+

When you dial 208-439-LUKE, your call goes into a live queue. Luke sees you waiting and can take your call right from the control room. Your voice streams in real time — no pre-recording, no delay. You're live on the show, talking to Luke, and the AI callers might even react to what you said. And if Luke isn't live, you can leave a voicemail — it gets transcribed and may get played on a future episode.

+
+
+ +
+
6
+
+

The Control Room

+

The entire show runs through a custom-built control panel. Luke manages callers, plays music and sound effects, runs ads, monitors the call queue, and controls everything from one screen. Audio is routed across multiple channels simultaneously — caller voices, music, sound effects, and live phone audio all on separate tracks. The website shows a live on-air indicator so listeners know when to call in.

+
+
+ Audio Channels + 5 independent +
+
+ Caller Slots + 10 per session +
+
+ Phone System + VoIP + WebSocket +
+
+ Live Status + Real-time CDN +
+
+
+
+
+
+
@@ -211,6 +314,12 @@
Social Clips +
+
+ +
+ Monitoring +
↓
@@ -281,108 +390,6 @@
- -
-

The Anatomy of an AI Caller

- -
-
-
1
-
-

A Person Is Born

-

Every caller starts as a blank slate. The system generates a complete identity: name, age, job, hometown, and personality. Each caller gets a unique speaking style — some ramble, some are blunt, some deflect with humor. They have relationships, vehicles, strong food opinions, nostalgic memories, and reasons for being up this late. They know what they were watching on TV, what errand they ran today, and what song was on the radio before they called.

-

Some callers become regulars. The system tracks returning callers across episodes — they remember past conversations, reference things they talked about before, and their stories evolve over time. You'll hear Carla update you on her divorce, or Carl check in about his gambling recovery. They're not reset between shows.

-
-
- Unique Names - 160 names -
-
- Personality Layers - 30+ -
-
- Towns with Real Knowledge - 32 -
-
- Returning Regulars - 12+ callers -
-
-
-
- -
-
2
-
-

They Know Their World

-

Callers know real facts about where they live — the restaurants, the highways, the local gossip. When a caller says they're from Lordsburg, they actually know about the Shakespeare ghost town and the drive to Deming. They know the current weather outside their window, what day of the week it is, whether it's monsoon season or chile harvest. They have strong opinions about where to get the best green chile and get nostalgic about how their town used to be. The system also pulls in real-time news so callers can reference things that actually happened today.

-
-
- -
-
3
-
-

They Have a Reason to Call

-

Some callers have a problem — a fight with a neighbor, a situation at work, something weighing on them at 2 AM. Others call to geek out about Severance, argue about poker strategy, or share something they read about quantum physics. Every caller has a purpose, not just a script.

-
-
- 70% - Need advice -
-
- 30% - Want to talk about something -
-
-
-
- -
-
4
-
-

The Conversation Is Real

-

Luke talks to each caller using push-to-talk, just like a real radio show. His voice is transcribed in real time, sent to an AI that responds in character, and then converted to speech using a voice engine — all in a few seconds. The AI doesn't just answer questions; it reacts, gets emotional, goes on tangents, and remembers what was said earlier in the show. Callers even react to previous callers — "Hey Luke, I heard that guy Tony earlier and I got to say, he's full of it." It makes the show feel like a living community, not isolated calls.

-
-
- -
-
5
-
-

Real Callers Call In Too

-

When you dial 208-439-LUKE, your call goes into a live queue. Luke sees you waiting and can take your call right from the control room. Your voice streams in real time — no pre-recording, no delay. You're live on the show, talking to Luke, and the AI callers might even react to what you said.

-
-
- -
-
6
-
-

The Control Room

-

The entire show runs through a custom-built control panel. Luke manages callers, plays music and sound effects, runs ads, monitors the call queue, and controls everything from one screen. Audio is routed across multiple channels simultaneously — caller voices, music, sound effects, and live phone audio all on separate tracks. The website shows a live on-air indicator so listeners know when to call in.

-
-
- Audio Channels - 5 independent -
-
- Caller Slots - 10 per session -
-
- Phone System - VoIP + WebSocket -
-
- Live Status - Real-time CDN -
-
-
-
-
-
-

From Live Show to Podcast

@@ -444,11 +451,11 @@
9

Automated Publishing

-

A single command takes a finished episode and handles everything: the audio is transcribed using speech recognition to generate full-text transcripts, then an LLM analyzes the transcript to write the episode title, description, and chapter markers with timestamps. The episode is uploaded to the podcast server, chapters and transcripts are attached to the metadata, and all media is synced to a global CDN so listeners everywhere get fast downloads.

+

A single command takes a finished episode and handles everything: the audio is transcribed using MLX Whisper running on Apple Silicon GPU to generate full-text transcripts, then an LLM analyzes the transcript to write the episode title, description, and chapter markers with timestamps. The episode is uploaded to the podcast server, chapters and transcripts are attached to the metadata, and all media is synced to a global CDN so listeners everywhere get fast downloads.

Transcription - Whisper AI + MLX Whisper (GPU)
Metadata @@ -470,7 +477,7 @@
10

Automated Social Clips

-

No manual editing, no scheduling tools. After each episode, an LLM reads the full transcript and picks the best moments — funny exchanges, wild confessions, heated debates. Each clip is automatically extracted, captioned with word-level timing, and rendered as a vertical video with the show's branding. A second LLM pass writes platform-specific descriptions and hashtags. Then a single script blasts every clip to Instagram Reels, YouTube Shorts, Facebook, Bluesky, and Mastodon simultaneously — six platforms, zero manual work.

+

No manual editing, no scheduling tools. After each episode, an LLM reads the full transcript and picks the best moments — funny exchanges, wild confessions, heated debates. Each clip is automatically extracted, transcribed with word-level timestamps, then polished by a second LLM pass that fixes punctuation, capitalization, and misheard words while preserving timing. The clips are rendered as vertical video with speaker-labeled captions and the show's branding. A third LLM writes platform-specific descriptions and hashtags. Then clips are uploaded directly to YouTube Shorts and Bluesky via their APIs, and pushed to Instagram Reels, Facebook, and Mastodon — six platforms, zero manual work.

Human Effort @@ -482,7 +489,7 @@
Captions - Word-level sync + LLM-polished
Simultaneous Push @@ -576,7 +583,7 @@
Or call in live: 208-439-LUKE
- Support the show on Ko-fi + Support the Show
@@ -602,9 +609,9 @@ YouTube -

Support the show: Ko-fi

+

Support the Show

Sales & Collaboration: luke@macneilmediagroup.com

-

© 2026 Luke at the Roost · Privacy Policy

+

© 2026 Luke at the Roost · Privacy Policy · System Status

diff --git a/website/index.html b/website/index.html index e67fc75..3ec1e37 100644 --- a/website/index.html +++ b/website/index.html @@ -117,12 +117,20 @@ · Discord · - Support the Show + Support the Show + +
+

Episodes

+
+
Loading episodes...
+
+
+

What Callers Are Saying

@@ -197,14 +205,6 @@
- -
-

Episodes

-
-
Loading episodes...
-
-
- diff --git a/website/privacy.html b/website/privacy.html index 4373b8a..8e40070 100644 --- a/website/privacy.html +++ b/website/privacy.html @@ -106,9 +106,9 @@ YouTube -

Support the show: Ko-fi

+

Support the Show

Sales & Collaboration: luke@macneilmediagroup.com

-

© 2026 Luke at the Roost · Privacy Policy

+

© 2026 Luke at the Roost · Privacy Policy · System Status

diff --git a/website/stats.html b/website/stats.html index 4a92f99..396b73d 100644 --- a/website/stats.html +++ b/website/stats.html @@ -78,9 +78,9 @@ YouTube -

Support the show: Ko-fi

+

Support the Show

Sales & Collaboration: luke@macneilmediagroup.com

-

© 2026 Luke at the Roost · Privacy Policy

+

© 2026 Luke at the Roost · Privacy Policy · System Status