Skip to content

API.md

Latest commit

 

History

History
30 lines (23 loc) · 1.86 KB

API.md

File metadata and controls

30 lines (23 loc) · 1.86 KB
Raw

TTSapi HTTP contract

The application uses multipart form fields; /docs supplies runtime OpenAPI after dependencies are installed. This table is source-derived, not a claim that an inference server was exercised here.

Method Path Auth Contract / limitation
POST /auth/token Username/password form Returns JWT; existing token generation has no expiry claim
POST /v1/tts Bearer Text, model, diffusion settings, format; optional voice/reference audio; audio response
POST /v1/jobs Bearer Same input; returns ID and queued status; state in memory
GET /v1/voices Bearer Lists local presets
POST /v1/voices Bearer Name, model and reference audio; saves style preset
DELETE /v1/voices/{name} Bearer Removes local preset; path/authorization hardening remains
POST /v1/bench Bearer Runs inference timings; consumes model compute
GET /v1/bench/public None Also runs inference; do not expose unrestricted
GET /healthz None Status/device detection; not proof that all model weights are loaded

Instrumentator may add /metrics at startup. Output formats: WAV, MP3, FLAC, OGG. Models: upstream LJSpeech and LibriTTS. Sample-rate encoding is 24 kHz in the current service.

Example local call

After configuring credentials and obtaining a token, set TOKEN locally:

curl --fail http://127.0.0.1:8000/v1/tts \
  -H "Authorization: Bearer $TOKEN" \
  -F 'text=Behind every line of code, there is a builder.' \
  -F 'model=LJSpeech' -F 'output_format=wav' --output speech.wav

No status/result retrieval endpoint exists for /v1/jobs. Request/voice ownership, parameter bounds, upload sizes, cancellation and durable state require further implementation. StreamingResponse transports completed encoded bytes; it is not incremental neural audio generation.