Ganas UK STT integration

Endpoint: https://api.uk.stt.ganas.ai

Authentication

Create a key in /admin. Send Authorization: Bearer YOUR_KEY. Keep keys on your backend; do not put them in public browser code.

Transcribe

curl --fail-with-body https://api.uk.stt.ganas.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $GANAS_STT_KEY" \
  -F file=@speech.wav -F language=English

{"text":"Good morning. How can I help?","language":"English"}

WAV/FLAC, mono or stereo, 8–48 kHz, 0.1–30 seconds. Maximum request body 10 MiB. Language: English (default), en, or auto. Audio is processed in memory and not retained.

Behaviour and limits

Whole-utterance transcription using Qwen3-ASR-1.7B with NF4 quantization on CUDA. This endpoint does not provide incremental streaming or timestamps. Quantized accuracy must be evaluated with your own UK-English calls.

Only one STT or TTS inference can run at a time on the shared GPU. On HTTP 429, retry after the Retry-After interval with jitter. No automatic retries of completed requests. HTTP 401 means missing/revoked key; 413 oversized upload; 422 invalid input; 503 model unavailable or inference failed.

Operations

/healthz: process liveness; /readyz: model loaded; /openapi.json: API schema (Bearer key required). Administrator credentials are separate from API keys. Usage statistics retain counts, key IDs, timestamps and latency for up to 30 days; request audio and text are not stored.