Skip to content

Configuration

Configuring the Engine

export ASR_ENGINE=openai_whisper

export ASR_ENGINE=faster_whisper

export ASR_ENGINE=whisperx

Configuring the Model

export ASR_MODEL=base

Available ASR_MODELs are:

  • Standard models: tiny, base, small, medium, large-v1, large-v2, large-v3 (or large), large-v3-turbo (or turbo)
  • English-optimized models: tiny.en, base.en, small.en, medium.en
  • Distilled models: distil-large-v2, distil-medium.en, distil-small.en, distil-large-v3 (only for whisperx and faster-whisper)

For English-only applications, the .en models tend to perform better, especially for the tiny.en and base.en models. We observed that the difference becomes less significant for the small.en and medium.en models.

The distilled models offer improved inference speed while maintaining good accuracy.

Configuring the Model Path

export ASR_MODEL_PATH=/data/whisper

Configuring the Model Unloading Timeout

export MODEL_IDLE_TIMEOUT=300

Defaults to 0. After no activity for this period (in seconds), unload the model until it is requested again. Setting 0 disables the timeout, keeping the model loaded indefinitely.

Configuring the SAMPLE_RATE

export SAMPLE_RATE=16000

Defaults to 16000. Default sample rate for audio input. 16 kHz is commonly used in speech-to-text tasks.

Configuring Device and Quantization

export ASR_DEVICE=cuda  # or 'cpu'
export ASR_QUANTIZATION=float32  # or 'float16', 'int8'

The ASR_DEVICE defaults to cuda if GPU is available, otherwise cpu.

The ASR_QUANTIZATION defines the precision for model weights:

  • float32: 32-bit floating-point precision (higher precision, slower inference)
  • float16: 16-bit floating-point precision (lower precision, faster inference)
  • int8: 8-bit integer precision (lowest precision, fastest inference)

Defaults to float32 for GPU, int8 for CPU.

Configuring Subtitle Options (WhisperX)

export SUBTITLE_MAX_LINE_WIDTH=1000
export SUBTITLE_MAX_LINE_COUNT=2
export SUBTITLE_HIGHLIGHT_WORDS=false

These options only apply when using the WhisperX engine:

  • SUBTITLE_MAX_LINE_WIDTH: Maximum width of subtitle lines (default: 1000)
  • SUBTITLE_MAX_LINE_COUNT: Maximum number of lines per subtitle (default: 2)
  • SUBTITLE_HIGHLIGHT_WORDS: Enable word highlighting in subtitles (default: false)

Hugging Face Token

export HF_TOKEN=your_token_here

Required when using the WhisperX engine to download the diarization model.