Sirene

Appearance

API Inference (FastAPI) ​

The inference service runs on port 8000 and handles all TTS model loading and processing. It is an internal API called only by the Hono server.

Generate Audio ​

http
POST /generate

Body:

json
{
  "backend": "kokoro",
  "text": "Hello, world!",
  "voice_path": "/data/models/kokoro/v1.0/voices/af_heart.pt",
  "speed": 1.0,
  "language": "en",
  "reference_audio": null,
  "reference_text": null,
  "model_path": "/data/models/kokoro/v1.0"
}

Response: audio/wav

Generate Audio (Streaming) ​

http
POST /generate/stream

Same body as above.

Response: audio/pcm chunked transfer encoding.

Backends ​

List Available Backends ​

http
GET /backends

Returns the list of registered TTS backends.

Get Backend Status ​

http
GET /backends/:name/status

Returns the backend's status including whether a model is loaded and memory usage.

Models ​

Pull a Model ​

http
POST /models/pull

Body:

json
{
  "url": "https://huggingface.co/...",
  "dest_path": "/data/models/kokoro/v1.0",
  "manifest": { }
}

Response: Server-Sent Events stream:

data: {"progress": 45, "speed": "12.5 MB/s", "status": "downloading"}
data: {"progress": 100, "status": "complete"}

Unload a Model ​

http
POST /models/unload

Body:

json
{
  "backend": "kokoro",
  "model_path": "/data/models/kokoro/v1.0"
}

Frees GPU memory by unloading the specified model.

Transcription ​

Transcribe Audio ​

http
POST /transcribe

Body: multipart/form-data with audio file.

Response:

json
{
  "text": "transcribed text",
  "language": "en"
}

Uses Whisper for automatic speech recognition.