Real-Time Speech Transcription

A real-time speech transcription system built with OpenAI Whisper supporting continuous audio recording and multiple languages.

Features

• Real-time transcription with continuous audio recording
• Multi-language support with auto-detection
• Segmented processing with smart overlap handling
• Interactive setup or command-line configuration
• Multiple recording methods: sounddevice, pyaudio, speech_recognition
• Pause/Resume functionality with Ctrl+C menu
• Audio segment saving for debugging and quality analysis
• Energy threshold configuration for speech detection

Installation

Prerequisites

• Python 3.8 or higher
• Conda (recommended) or pip

Create Environment

Using Conda (Recommended):

# Create and activate conda environment
conda create -n whisper-transcription python=3.9
conda activate whisper-transcription

# Clone the repository
git clone https://github.com/lancer1911/real-time-whisper-transcription.git
cd real-time-whisper-transcription

# Install dependencies
pip install -r requirements.txt

# Optional: Install additional packages
pip install sounddevice faster-whisper

Using pip:

# Clone the repository
git clone https://github.com/lancer1911/real-time-whisper-transcription.git
cd real-time-whisper-transcription

# Install dependencies
pip install -r requirements.txt

# Optional: Install additional packages
pip install sounddevice faster-whisper

Note: The requirements.txt includes PyTorch with CUDA 11.6 support. For CPU-only installation, modify the torch installation:

pip install torch --index-url https://download.pytorch.org/whl/cpu

System Dependencies

macOS:

# Install portaudio for pyaudio
brew install portaudio

Ubuntu/Debian:

# Install portaudio and ffmpeg
sudo apt update
sudo apt install portaudio19-dev ffmpeg

Windows:

• Download and install FFmpeg from https://ffmpeg.org/download.html
• Add FFmpeg to PATH environment variable

Quick Start

Activate Environment

# If using conda
conda activate whisper-transcription

Interactive Mode

python transcriber.py

Command Line Mode

# English transcription
python transcriber.py \                                       
  --language en \
  --model small \
  --save_audio \
  --output_file transcription.txt \
  --device cpu \
  --microphone_index 4 \
  --energy_threshold 100 \
  --segment_duration 8.0 \
  --overlap_duration 1.0 \
  --debug

# Chinese transcription
python transcriber.py \                                       
  --language zh \
  --model small \
  --save_audio \
  --output_file transcription.txt \
  --device cpu \
  --microphone_index 4 \
  --energy_threshold 100 \
  --segment_duration 8.0 \
  --overlap_duration 1.0 \
  --debug

Audio Diagnostics

Test your microphone setup and find optimal settings:

python audio_diagnostic.py

The diagnostic tool will:

• List all available microphone devices
• Test different energy thresholds (4000, 1000, 300, 100, 50)
• Analyze audio volume levels
• Provide recommended energy threshold values

Sample output:

1. Available microphone devices:
   0: Built-in Microphone
   1: USB Audio Device

2. Testing default microphone...
   Testing energy threshold: 300
   Volume - Max: 0.156, Average: 0.023
   Threshold 300 working properly!

Recommended energy threshold: 300

Use the recommended settings:

python transcriber.py \                                       
  --language en \
  --model small \
  --save_audio \
  --output_file transcription.txt \
  --device cpu \
  --microphone_index 4 \
  --energy_threshold 100 \
  --segment_duration 8.0 \
  --overlap_duration 1.0 \
  --debug

Configuration

Basic Parameters

• --language: Target language (en, zh, ja, ko, es, fr, de, auto)
• --model: Whisper model size (tiny, base, small, medium, large)
• --microphone_index: Microphone device index
• --energy_threshold: Speech detection threshold (default: 300)

Audio Settings

• --recording_method: Recording method (sounddevice, pyaudio, speech_recognition)
• --segment_duration: Segment length in seconds (default: 8.0)
• --overlap_duration: Overlap length in seconds (default: 2.0)

Output Settings

• --output_file: Transcription output file (default: transcription.txt)
• --save_audio: Save audio segments (recommended for debugging and analysis)
• --debug: Enable debug mode

Interactive Controls

During transcription:

• Ctrl+C: Pause and access options menu
  • Option 1: Resume recording
  • Option 2: Stop and exit

Requirements

Core Dependencies (in requirements.txt)

• Python 3.8+
• PyTorch (with CUDA 11.6 support)
• OpenAI Whisper
• NumPy
• PyAudio
• SpeechRecognition

Optional Dependencies

• sounddevice (recommended for better audio recording)
• faster-whisper (experimental performance improvements)

System Requirements

• 4GB RAM minimum
• Microphone access
• Internet connection for initial model download
• CUDA-capable GPU (optional, for acceleration)

License

MIT License - see LICENSE file for details.