STTClientConfig
===============

**class** ``STTClientConfig`` **(** :doc:`BaseClientConfig` **)**

    STT client configuration for realtime streaming speech-to-text APIs.

**Polymorphic Type:**

    ``type: stt``

    All ``BaseClientConfig`` types:

    - ``openai_chat_completions``: :doc:`OpenAIChatCompletionsClientConfig`
    - ``tts``: :doc:`TTSClientConfig`
    - ``streaming_tts``: :doc:`StreamingTTSClientConfig`
    - ``stt``: :doc:`STTClientConfig`
    - ``openai_completions``: :doc:`OpenAICompletionsClientConfig`
    - ``openai_router``: :doc:`OpenAIRouterClientConfig`

**Fields:**

``api_base`` : *Optional* [ *str* ] = ``None``
    API base URL. Defaults to OPENAI_API_BASE env var.

``api_key`` : *Optional* [ *str* ] = ``None``
    API key. Defaults to OPENAI_API_KEY env var.

``model`` : *str* = ``""``
    The STT model ID.

``address_append_value`` : *str* = ``"chat/completions"``
    The address append value for the LLM API.

``request_timeout`` : *int* = ``300``
    The timeout for each request to the LLM API (in seconds).

``additional_sampling_params`` : *str* = ``"{}"``
    Additional sampling params to send with each request to the LLM API.

``provider`` : *str* = ``""``
    STT provider name. Supported: vajra_openai_realtime, vllm_realtime, deepgram_flux, deepgram_nova, elevenlabs, mistral, cartesia, together.

``sample_rate`` : *int* = ``16000``
    Expected audio sample rate in Hz.

``ws_chunk_size`` : *int* = ``4096``
    Bytes of raw PCM audio per WebSocket message. Client CPU scales with concurrency * sample_rate * 2 / ws_chunk_size, so prefer larger chunks at high concurrency.

``ws_permessage_deflate`` : *bool* = ``False``
    Negotiate WebSocket permessage-deflate compression. Disabled by default because base64 PCM is high entropy and compression adds substantial client and server CPU.

``ws_realtime_pacing`` : *bool* = ``False``
    Sleep between WebSocket audio chunks to simulate realtime input. Enable for live-audio SLO measurements; disable for engine-bound throughput measurements.

``ws_ping_interval_s`` : *Optional* [ *int* ] = ``20``
    WebSocket ping interval in seconds; None disables pings.

``ws_ping_timeout_s`` : *Optional* [ *int* ] = ``None``
    WebSocket ping timeout in seconds. None disables keepalive timeout while preserving request_timeout.

``api_key_env`` : *Optional* [ *str* ] = ``None``
    Optional provider API-key environment variable override.

``language`` : *str* = ``"en"``
    Requested transcription language.

``mip_opt_out`` : *bool* = ``False``
    Opt out of the Deepgram Model Improvement Program.

``target_streaming_delay_ms`` : *Optional* [ *int* ] = ``None``
    Optional Mistral transcription context delay in milliseconds.

``cartesia_version`` : *str* = ``"2026-03-01"``
    Cartesia API version header.