dinscribe 0.2.0

Audio denoising and transcription pipeline using Demucs, Silero VAD, and Whisper

dinscribe audio transcription

Processes audio through a three-step pipeline to produce a transcription JSON: denoising (demucs), voice activity detection (Silero VAD), and transcription (Whisper).

Installation

pip install dinscribe

This installs CPU-only torch by default. For GPU acceleration, install the CUDA build first:

pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu124
pip install dinscribe

dinscribe uses CUDA automatically when available and warns at startup if it is not.

On first run, dinscribe copies default config files to your platform config directory:

• Windows: %APPDATA%\dinscribe\
• macOS: ~/Library/Application Support/dinscribe/
• Linux: ~/.config/dinscribe/

Edit config.yaml and vocab.txt to customize settings.

CLI usage

dinscribe input/audio.mp3          # single file
dinscribe input/                   # all audio files in a folder
dinscribe input/audio.mp3 -f       # force re-run all steps
dinscribe input/audio.mp3 -c path/to/config.yaml   # custom config
dinscribe input/audio.mp3 -o results/              # custom output dir

Each step checks whether its output already exists and skips it if so. Use -f to force all steps to re-run.

Output is written to output/<filename>/ and contains:

• <filename>_denoised.wav (vocals isolated from background noise)
• <filename>_vad.json (detected speech segment boundaries)
• <filename>_transcription.json (final transcription with timestamps)

Python API

from pathlib import Path
import dinscribe
from dinscribe import PipelineConfig, VadConfig, TranscribeConfig

# Run the full pipeline with defaults
dinscribe.process_file(
    input_path=Path("recording.wav"),
    output_dir=Path("output"),
)

# Custom config
config = PipelineConfig(
    vad=VadConfig(threshold=0.4, max_segment_length_sec=20),
    transcribe=TranscribeConfig(model="small", language="en"),
)
dinscribe.process_file(Path("recording.wav"), Path("output"), config=config)

# Or use individual stages
from dinscribe import denoise, vad, transcribe

denoised = denoise.run(Path("recording.wav"), Path("output/recording"))
vad_file  = vad.run(denoised, Path("output/recording"))
result    = transcribe.run(denoised, vad_file, Path("output/recording"))

Configuration

denoise:
  model: htdemucs        # htdemucs | htdemucs_ft | mdx | mdx_extra | htdemucs_6s

vad:
  threshold: 0.5         # 0.0–1.0, higher = requires clearer speech
  min_speech_duration_ms: 250
  min_silence_duration_ms: 100
  padding_ms: 500
  max_segment_length_sec: 30
  merge_within_sec: 1.0

transcribe:
  model: base            # tiny | base | small | medium | large
  language: en           # set to null to auto-detect
  temperature: null      # null = Whisper fallback sequence, 0 = greedy
  no_speech_threshold: 0.6
  logprob_threshold: -1.0
  compression_ratio_threshold: 2.4
  condition_on_previous_text: false
  vocab_file: null       # path to domain-specific vocabulary, defaults to vocab.txt in config dir

Add domain-specific vocabulary to vocab.txt to improve transcription accuracy on unusual words and jargon. For noisy or technical audio, set temperature: 0 to disable attempts to fallback to higher-temperature decoding, and consider filtering out any common hallucinations specific to your dataset.