REM · STT
Speech-to-text
For the metered Rem stream, create a session at https://app.lucidweights.ai/api/v1/stt/sessions. The older batch and streaming endpoints remain at https://api.lucidweights.ai/v1/stt/; they are a separate service.
Endpoints
| Method | Path | Purpose |
|---|---|---|
| POST | app.lucidweights.ai/api/v1/stt/sessions | Create a metered Rem stream with an lw_ key |
| WS | Returned wsUrl | Send 16 kHz mono PCM16 and receive transcripts |
| POST | /v1/stt/load_model | Older STT service: warm a backend |
| POST | /v1/stt/transcribe | Older STT service: transcribe base64 raw audio |
| POST | /v1/stt/file-transcribe | Older STT service: transcribe an uploaded file |
| WS | /v1/stt/stream | Older STT service: separate live stream |
Older STT service backends
ultrafastnormal
ultra is the default and recommended backend: fastest, auto-detects language, and model_name does not apply. fast and normal are multilingual and size-selectable.
Model names (fast / normal)
tinybasesmallmediumlarge-v2large-v3largelarge-v3-turboturbo
Languages query param
| Value | Behaviour |
|---|---|
| omitted | Full auto-detect |
| one code: en | Language forced; lowest latency |
| several: en,es,hi | fast/normal pick best match; ultra ignores multi-language hints |