# SmartMeeting

SmartMeeting is a Docker Compose deployment for Speakr, FunASR Paraformer ASR,
speaker embedding extraction, and the realtime WebSocket bridge.

## Directory Layout

```text
/storage01/home/xll/202607/voice
├── docker-compose.yml          # Main Compose file
├── config.env                  # Runtime configuration to edit
├── asr/main.py                 # ASR adapter service code
├── speakr/                     # Speakr frontend/backend code
├── voice-dected/               # Speaker embedding service code
├── fastapi_wss/                # Realtime WebSocket bridge code
└── deploy/data/                # Runtime data, ignored by git
```

## Start And Stop

Run all services:

```bash
cd /storage01/home/xll/202607/voice
docker compose up -d
```

Check status and logs:

```bash
docker compose ps
docker compose logs -f
```

Stop services:

```bash
docker compose down
```

## Services And Ports

| Service | Container | Port | Purpose |
| --- | --- | --- | --- |
| Speakr | `voice-speakr` | `8899` | Web UI and backend |
| ASR | `asr-test` | `59805` | Paraformer HTTP and WebSocket ASR |
| Speaker embedding | `voice-dected` | `18000` | `/extract_embedding` voiceprint service |
| Realtime bridge | `fastapi-wss` | `10095` | Browser realtime ASR WebSocket bridge |
| Redis | `voice-redis` | `16380` | Cache/session storage for realtime bridge |

Main access URL:

```text
http://192.168.0.46:8899/tool/speakr/
```

For browser microphone recording, use a secure context. The quickest local
workaround is an SSH tunnel:

```bash
ssh -L 8899:127.0.0.1:8899 xll@192.168.0.46
```

Then open:

```text
http://localhost:8899/tool/speakr/
```

## Configuration

Most runtime settings are in:

```text
/storage01/home/xll/202607/voice/config.env
```

Edit this file first when changing service addresses, model endpoints, LLM
settings, or speaker matching behavior. After editing:

```bash
cd /storage01/home/xll/202607/voice
docker compose up -d
```

Important variables:

```env
ASR_BASE_URL=http://asr-test:59805
TARGET_WS_URL=ws://asr-test:59805/ws/asr
VOICE_DETECTED_URL=http://voice-dected:8000
SPEAKER_DET_URL=http://voice-dected:8000/extract_embedding
REDIS_URL=redis://redis:6379/0

TEXT_MODEL_BASE_URL=http://192.168.0.46:59800/v1
TEXT_MODEL_NAME=46-qwen3.5-35B
TEXT_MODEL_API_KEY=none

VOICEPRINT_STORE=/data/asr_voiceprints.json
VOICEPRINT_MATCH_THRESHOLD=0.45
```

Notes:

- Inside Docker Compose, use service names such as `asr-test`,
  `voice-dected`, and `redis`.
- From outside Docker, use host ports such as `192.168.0.46:59805`,
  `192.168.0.46:18000`, and `192.168.0.46:10095`.
- `VOICEPRINT_MATCH_THRESHOLD` controls how strict speaker matching is.
  Lower values match more easily; higher values reduce false matches.

## Runtime Data

Runtime data is stored under:

```text
/storage01/home/xll/202607/voice/deploy/data/
```

Important subdirectories:

```text
deploy/data/speakr/uploads/      # Uploaded/recorded audio
deploy/data/speakr/instance/     # Speakr SQLite database
deploy/data/asr/                 # ASR voiceprint store
deploy/data/redis/               # Redis persistence
```

These files are ignored by git.

## Voiceprint Flow

Speakr calls the ASR adapter for voiceprint platform compatibility:

```text
POST   /voice_insert
GET    /get_voice_name
DELETE /delete_voice
```

The ASR adapter stores embeddings in:

```text
deploy/data/asr/asr_voiceprints.json
```

During `/asr` transcription, the adapter converts uploaded audio to 16 kHz mono
WAV, calls `voice-dected`, compares the embedding with stored voiceprints, and
returns a matched speaker name when the similarity passes the threshold.

## Useful Checks

```bash
curl http://127.0.0.1:59805/health
curl http://127.0.0.1:59805/get_voice_name
curl http://127.0.0.1:8899/tool/speakr/api/config
```

WebSocket check:

```bash
docker exec -it fastapi-wss python - <<'PY'
import asyncio, json, websockets

async def main():
    async with websockets.connect("ws://127.0.0.1:10095") as ws:
        await ws.send(json.dumps({"mode": "2pass", "is_speaking": True}))
        await ws.send(json.dumps({"type": "stop", "is_speaking": False}))
        print(await ws.recv())

asyncio.run(main())
PY
```