tcpsyn
7b7f9b8208
- On-air toggle uploads status.json to BunnyCDN + purges cache, website polls it every 15s to show live ON AIR / OFF AIR badge - Publish script downloads Castopod's copy of audio for CDN upload (byte-exact match), removes broken slug fallback, syncs all episode media to CDN after publishing - Fix f-string syntax error in publish_episode.py (Python <3.12) - Enable CORS on BunnyCDN pull zone for json files - CDN URLs for website OG images, stem recorder bug fixes, LLM token budget tweaks, session context in CLAUDE.md Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
66 lines
3.3 KiB
Markdown
66 lines
3.3 KiB
Markdown
# AI Podcast - Project Instructions
|
|
|
|
## Git Remote (Gitea)
|
|
- **Repo**: `git@gitea-nas:luke/ai-podcast.git`
|
|
- **Web**: http://mmgnas:3000/luke/ai-podcast
|
|
- **SSH Host**: `gitea-nas` (configured in ~/.ssh/config)
|
|
- HostName: `mmgnas` (use `mmgnas-10g` if wired connection issues)
|
|
- Port: `2222`
|
|
- User: `git`
|
|
- IdentityFile: `~/.ssh/gitea_mmgnas`
|
|
|
|
## NAS Access
|
|
- **Hostname**: `mmgnas` (wireless) or `mmgnas-10g` (wired/10G)
|
|
- **SSH Port**: 8001
|
|
- **User**: luke
|
|
- **Docker path**: `/share/CACHEDEV1_DATA/.qpkg/container-station/bin/docker`
|
|
|
|
## Castopod (Podcast Publishing)
|
|
- **URL**: https://podcast.macneilmediagroup.com
|
|
- **Podcast handle**: `@LukeAtTheRoost`
|
|
- **API Auth**: Basic auth (admin/podcast2026api)
|
|
- **Container**: `castopod-castopod-1`
|
|
- **Database**: `castopod-mariadb-1` (user: castopod, db: castopod)
|
|
|
|
## Running the App
|
|
```bash
|
|
# Start backend
|
|
cd /Users/lukemacneil/ai-podcast
|
|
python -m uvicorn backend.main:app --reload --host 0.0.0.0 --port 8000
|
|
|
|
# Or use run.sh
|
|
./run.sh
|
|
```
|
|
|
|
## Publishing Episodes
|
|
```bash
|
|
python publish_episode.py ~/Desktop/episode.mp3
|
|
```
|
|
|
|
## Environment Variables
|
|
Required in `.env`:
|
|
- OPENROUTER_API_KEY
|
|
- ELEVENLABS_API_KEY (optional)
|
|
- INWORLD_API_KEY (for Inworld TTS)
|
|
|
|
## Post-Production Pipeline (added Feb 2026)
|
|
- **Branch**: `feature/real-callers` — all current work is here, pushed to gitea
|
|
- **Stem Recorder** (`backend/services/stem_recorder.py`): Records 5 WAV stems (host, caller, music, sfx, ads) during live shows. Uses lock-free deque architecture — audio callbacks just append to deques, a background writer thread drains to disk. `write()` for continuous streams (host mic, music, ads), `write_sporadic()` for burst sources (caller TTS, SFX) with time-aligned silence padding.
|
|
- **Audio hooks** in `backend/services/audio.py`: 7 tap points guarded by `if self.stem_recorder:`. Persistent mic stream (`start_stem_mic`/`stop_stem_mic`) runs during recording to capture host voice continuously, not just during push-to-talk.
|
|
- **API endpoints**: `POST /api/recording/start`, `POST /api/recording/stop` (auto-runs postprod in background thread), `POST /api/recording/process`
|
|
- **Frontend**: REC button in header with red pulse animation when recording
|
|
- **Post-prod script** (`postprod.py`): 6-step pipeline — load stems → gap removal → voice compression (ffmpeg acompressor) → music ducking → stereo mix → EBU R128 loudness normalization to -16 LUFS. All steps skippable via CLI flags.
|
|
- **Known issues resolved**: Lock-free recorder (old version used threading.Lock in audio callbacks causing crashes), scipy.signal.resample replaced with nearest-neighbor (was producing artifacts on small chunks), sys import bug in auto-postprod, host mic not captured without persistent stream
|
|
|
|
## LLM Settings
|
|
- `_pick_response_budget()` in main.py controls caller dialog token limits (150-450 tokens). MiniMax respects limits strictly — if responses seem short, check these values.
|
|
- Default max_tokens in llm.py is 300 (for non-caller uses)
|
|
- Grok (`x-ai/grok-4-fast`) works well for natural dialog; MiniMax tends toward terse responses
|
|
|
|
## Website
|
|
- **Domain**: lukeattheroost.com (behind Cloudflare)
|
|
- **Analytics**: Cloudflare Web Analytics (enable in Cloudflare dashboard, no code changes needed)
|
|
|
|
## Episodes Published
|
|
- Episode 6 published 2026-02-08 (podcast6.mp3, ~31 min)
|