92 lines
2.8 KiB
Markdown
92 lines
2.8 KiB
Markdown
# FRIDAY Voice Console
|
|
|
|
Push-to-talk MVP for talking to FRIDAY from a browser/phone, with a speaker-identification foundation for Jeremy's voice.
|
|
|
|
## Status
|
|
|
|
MVP scaffold complete:
|
|
|
|
- Browser UI records audio and sends it to the backend.
|
|
- Backend transcribes through Speaches on the GPU machine.
|
|
- Backend routes text into Hermes via `hermes chat -q`.
|
|
- Backend speaks FRIDAY's response through Speaches/Kokoro.
|
|
- JSON-backed speaker profile store supports voice embeddings and confidence thresholds.
|
|
- Unknown speakers are treated as read-only.
|
|
|
|
## Runtime defaults
|
|
|
|
- Speaches: `http://192.168.86.150:8000`
|
|
- Hermes source: `friday-voice`
|
|
- Confirmed speaker threshold: `0.85`
|
|
- Probable speaker threshold: `0.70`
|
|
- Data directory in container: `/data`
|
|
|
|
## Local verification
|
|
|
|
```bash
|
|
cd /mnt/c/Users/Jeremy/projects/friday-voice
|
|
python3 -m unittest discover -s tests -v
|
|
python3 -m compileall friday_voice
|
|
node --check web/app.js
|
|
```
|
|
|
|
## Deployment layout
|
|
|
|
Source project:
|
|
|
|
```text
|
|
/mnt/c/Users/Jeremy/projects/friday-voice/
|
|
```
|
|
|
|
Compose definition:
|
|
|
|
```text
|
|
/mnt/c/Users/Jeremy/projects/docker/compose/hf-pdocker-01/friday-voice/docker-compose.yml
|
|
```
|
|
|
|
Traefik dynamic config draft:
|
|
|
|
```text
|
|
/mnt/c/Users/Jeremy/projects/docker/stacks/traefik/dynamic/friday-voice.yml
|
|
```
|
|
|
|
Target host layout:
|
|
|
|
```text
|
|
/opt/friday-voice/ # application source/build context
|
|
/mnt/docker-data/compose/hf-pdocker-01/friday-voice/
|
|
/mnt/docker-data/configs/friday-voice/ # speaker profiles + generated TTS cache
|
|
/mnt/docker-data/traefik/dynamic/friday-voice.yml
|
|
```
|
|
|
|
Deploy protocol, after review:
|
|
|
|
```bash
|
|
# From workstation
|
|
rsync -av --delete /mnt/c/Users/Jeremy/projects/friday-voice/ glyptic@192.168.86.192:/opt/friday-voice/
|
|
rsync -av /mnt/c/Users/Jeremy/projects/docker/compose/hf-pdocker-01/friday-voice/ glyptic@192.168.86.192:/mnt/docker-data/compose/hf-pdocker-01/friday-voice/
|
|
rsync -av /mnt/c/Users/Jeremy/projects/docker/stacks/traefik/dynamic/friday-voice.yml glyptic@192.168.86.192:/mnt/docker-data/traefik/dynamic/friday-voice.yml
|
|
|
|
# On hf-pdocker-01
|
|
cd /mnt/docker-data/compose/hf-pdocker-01/friday-voice
|
|
docker compose up -d --build
|
|
curl -fsS http://127.0.0.1:8320/health
|
|
```
|
|
|
|
## Speaker enrollment MVP
|
|
|
|
The current MVP exposes `/api/enroll` and `/api/identify` for embedding vectors. The next build slice should add actual audio-to-embedding extraction using SpeechBrain ECAPA or resemblyzer, then wire that into the browser enrollment flow.
|
|
|
|
Recommended thresholds:
|
|
|
|
- `>= 0.85`: confirmed Jeremy
|
|
- `0.70-0.84`: probable Jeremy, confirmation required for sensitive actions
|
|
- `< 0.70`: unknown speaker, read-only
|
|
|
|
## Security notes
|
|
|
|
- Voice identity is advisory, not a sole authorization factor.
|
|
- Unknown speakers are read-only.
|
|
- Environment writes still require explicit authorization and wiki documentation.
|
|
- Secrets are never returned by voice.
|