| # API reference |
|
|
| All URLs below are relative to your Morisyen TTS service URL. |
|
|
| ## Health |
|
|
| `GET /health` |
|
|
| ```json |
| {"status": "ok"} |
| ``` |
|
|
| This is a service liveness response. It does not confirm model readiness or audio quality. |
|
|
| ## Complete synthesis |
|
|
| `POST /synthesize` |
|
|
| Request header: `Content-Type: application/json`. |
|
|
| ```json |
| {"gen_text": "Bonzur! Kouma ou ete zordi?"} |
| ``` |
|
|
| `gen_text` is a required, non-empty string. Successful responses contain: |
|
|
| | Field | Meaning | |
| | --- | --- | |
| | `audio_base64` | Base64-encoded complete WAV file | |
| | `content_type` | `audio/wav` | |
| | `sample_rate` | `8000` Hz | |
| | `encoding` | `PCM_16` | |
| | `duration_seconds` | Generated audio duration in seconds | |
| | `byte_size` | WAV file length in bytes | |
|
|
| Decode `audio_base64` and save the bytes as a `.wav` file. The base64 text is not the audio file itself. |
|
|
| Invalid or missing input returns an HTTP error, commonly `400` or `422`. Synthesis failures return `500`. Authentication and access controls, if configured, depend on your deployment. |
|
|
| ## Streaming synthesis |
|
|
| `WS /ws/synthesize` |
|
|
| 1. Connect to the WebSocket endpoint. |
| 2. Send one JSON text frame containing `gen_text`. |
| 3. Consume binary frames; each frame is a complete mono PCM16 WAV chunk at 8,000 Hz. |
| 4. Read the final JSON summary and allow the connection to close. |
|
|
| An error is delivered as a JSON frame with an `error` field. Summary fields can vary by service version. Audio duration and request latency are different measurements. |
|
|
| For continuous playback, extract each WAV chunk's PCM frames before appending them to your audio buffer. Use `client.stream(text)` to perform that conversion automatically. |
|
|
