pyflink.multimodal.expression.AudioExpressionAccessor.asr_whisper

AudioExpressionAccessor.asr_whisper(*, language: Optional[str] = None, task: Literal['transcribe', 'translate'] = 'transcribe', model: str = 'openai/whisper-tiny', model_sharing: Optional[Literal['process', 'shared']] = None, concurrency: Optional[int] = None, batch_size: Optional[int] = None, num_gpus: Optional[float] = None, gpu_type: Optional[str] = None) → pyflink.table.expression.Expression

Run Whisper ASR over waveform values.

Equivalent to audio_asr_whisper(). See that function for full parameter details.

Parameters

• language – Optional source language hint.

• task – "transcribe" or "translate".

• model – Whisper or Whisper-compatible model id.

• model_sharing – Optional model sharing mode.

• concurrency – Optional execution concurrency for this operation. None uses the framework default.

• batch_size – Optional inference batch size.

• num_gpus – Optional number of GPUs requested for model inference.

• gpu_type – Optional GPU type requested for model inference.

Returns

A DataFrame expression that produces an ASR result ROW with asr_result, timestamps, and segments fields.