AudioTraceFlavorConfig
======================

**class** ``AudioTraceFlavorConfig`` **(** :doc:`BaseTraceFlavorConfig` **)**

    Audio trace flavor configuration for STT benchmarking.

**Polymorphic Type:**

    ``type: audio``

    All ``BaseTraceFlavorConfig`` types:

    - ``timed_synthetic_session``: :doc:`TimedSyntheticSessionTraceFlavorConfig`
    - ``shared_prefix``: :doc:`SharedPrefixTraceFlavorConfig`
    - ``rag``: :doc:`RAGTraceFlavorConfig`
    - ``request_log``: :doc:`RequestLogTraceFlavorConfig`
    - ``untimed_content_multi_turn``: :doc:`UntimedContentMultiTurnTraceFlavorConfig`
    - ``sharegpt``: :doc:`ShareGPTTraceFlavorConfig`
    - ``audio``: :doc:`AudioTraceFlavorConfig`
    - ``seed_tts_text``: :doc:`SeedTTSTextTraceFlavorConfig`

**Fields:**

``audio_dir`` : *str* = ``""``
    Optional base directory prepended to relative audio_file paths.

``target_duration_s`` : *Optional* [ *float* ] = ``None``
    Optional per-session audio duration to stream. The trace must provide reference_word_timestamps so expected transcripts can be trimmed to the streamed prefix.

``target_duration_spread_s`` : *Optional* [ *float* ] = ``None``
    Hard half-width of a clipped-Gaussian duration spread around target_duration_s. Requires target_duration_s.

``target_duration_sigma_s`` : *Optional* [ *float* ] = ``None``
    Standard deviation of the clipped-Gaussian duration draw. Defaults to half the spread and must not exceed the spread.