Usage guide
Setup, environment variables, and Python/JavaScript/Go usage examples.
OpenAISTT
Speech-to-text with OpenAI’s transcription models, streamed in real time or transcribed per turn. Good for accurate multilingual transcription with your own OpenAI key. Authenticates with OPENAI_API_KEY from your environment, or an explicitapi_key.
Fields
str | None
default:"UNSET"
Transcription model, e.g. “gpt-live-transcribe” for streaming or “gpt-transcribe” for per-turn. Leave unset to pick based on enable_streaming. Defaults to None (not set).
str | None
default:"UNSET"
Custom OpenAI API base URL, e.g. for a proxy. Defaults to None (not set).
str | None
default:"UNSET"
Text hint to guide transcription, such as expected vocabulary, names or spelling. Defaults to None (not set).
str
default:"UNSET"
ISO-639-1 language code of the speech, e.g. “en” or “es”. Defaults to ‘en’.
dict | None
default:"UNSET"
Server-side turn detection settings for streaming, e.g. {“type”: “server_vad”}. Newer models reject it and rely on the agent’s own VAD. Defaults to None (not set).
bool
default:"UNSET"
Stream audio in real time over a live connection instead of sending each turn as a single request. Defaults to True.
float
default:"UNSET"
Audio level, 0 to 1, below which input counts as silence when segmenting speech without streaming. Defaults to 0.01.
float
default:"UNSET"
Seconds of silence that end an utterance when segmenting speech without streaming. Defaults to 0.8.
bool
default:"True"
Whether partial transcripts are also sent to the client as the user speaks, not just the final one. Defaults to True.