whisper-transcriber 0.2.0

A library for transcribing audio files using Whisper models

Whisper Transcriber

A Python library for transcribing audio files using Whisper models with intelligent silence detection and segmentation.

Installation

pip install whisper-transcriber

Requirements

• Python 3.7 or higher
• ffmpeg and ffprobe installed on your system

Features

• Intelligent silence detection for natural segmentation
• Adaptive audio analysis for optimal threshold detection
• High-quality transcription using Whisper models
• Support for various audio formats
• Optional SRT subtitle output
• Control over transcript output (quiet mode, JSON output)
• Verbose/silent operation modes

Usage

Command Line

# Basic usage
whisper-transcribe audio_file.mp3

# Advanced usage
whisper-transcribe audio_file.mp3 -m openai/whisper-small \
  --min-segment 5 \
  --max-segment 15 \
  --silence-duration 0.2 \
  --sample-rate 16000 \
  --batch-size 8 \
  --normalize \
  --hf-token YOUR_HF_TOKEN \
  --no-timestamps

# Run in quiet mode (no transcript printing during processing)
whisper-transcribe audio_file.mp3 --quiet

# Output results as JSON
whisper-transcribe audio_file.mp3 --json

Available Arguments:

• input: Input audio file or directory (required)
• -o, --output: Output file path (optional)
• -m, --model: Whisper model to use (default: openai/whisper-small)
• --hf-token: HuggingFace API token
• --min-segment: Minimum segment length in seconds (default: 5)
• --max-segment: Maximum segment length in seconds (default: 15)
• --silence-duration: Minimum silence duration in seconds (default: 0.2)
• --sample-rate: Audio sample rate (default: 16000)
• --batch-size: Batch size for transcription (default: 8)
• --normalize: Normalize audio volume
• --no-text-normalize: Skip text normalization
• --no-timestamps: Don't print timestamps during processing
• --quiet: Run in quiet mode (suppress transcript printing)
• --json: Output results as JSON instead of text

Python Library

from whisper_transcriber import WhisperTranscriber

# Initialize the transcriber
transcriber = WhisperTranscriber(model_name="openai/whisper-small", hf_token="YOUR_HF_TOKEN")

# Transcribe an audio file with automatic transcript printing
results = transcriber.transcribe(
    "audio_file.mp3",
    min_segment=5,
    max_segment=15,
    silence_duration=0.2,
    sample_rate=16000,
    batch_size=8,
    normalize=True,
    normalize_text=True,
    print_timestamps=True,
    verbose=False
)



# Access the transcription results manually
for i, segment in enumerate(results):
    print(f"\n[{segment['start']} --> {segment['end']}]")
    print(f"Segment {i+1}: {segment['transcript']}")

# Optionally save to an SRT file
# If you want to save the transcription, provide an output path
results = transcriber.transcribe(
    "audio_file.mp3",
    output="transcript.srt"
)

Parameters Explained

• model_name: Which Whisper model to use (e.g., "openai/whisper-tiny", "openai/whisper-small", "openai/whisper-medium", "openai/whisper-large")
• min_segment: Minimum length in seconds for audio segments (shorter segments will be merged)
• max_segment: Maximum length in seconds for audio segments (longer segments will be split)
• silence_duration: How long a silence needs to be (in seconds) to be considered a natural break point
• sample_rate: Audio sample rate in Hz for processing
• batch_size: Number of segments to process at once (higher values use more memory but can be faster with GPU)
• normalize: Whether to normalize audio volume
• normalize_text: Whether to normalize transcription text
• print_timestamps: Whether to include timestamps when printing transcripts
• verbose: Whether to print processing information and transcripts during transcription

License

MIT