Enhanced Speech Clarity Tool

A tool for converting difficult-to-understand speech (such as accented English) into clear, pleasant voices with disfluency removal.

Features

• Extract audio from YouTube videos or local files
• Transcribe speech with high accuracy using OpenAI's Whisper (locally)
• Remove speech disfluencies (um, uh, false starts, repetitions)
• Convert to natural-sounding speech using high-quality neural TTS
• Maintain the original timing and pacing of the conversation

Requirements

• Python 3.8 or higher
• FFmpeg
• 4GB+ RAM for transcription
• 500MB+ disk space for models

Installation

Quick Setup (Recommended)

Run the provided setup script, which will install all required dependencies and download the necessary models:

# Make the setup script executable
chmod +x setup.sh

# Run the setup script
./setup.sh

Manual Installation

If you prefer to install manually:

1. Create a virtual environment:

   python3 -m venv .venv
   source .venv/bin/activate

2. Install Python dependencies:

   pip install yt-dlp pydub tqdm openai-whisper piper-tts torch

3. Download the TTS voice model:

   piper-download --voice en_US-lessac-medium

Usage

Process a YouTube Video

python enhanced_speech_tool.py -yt "https://www.youtube.com/watch?v=YOUR_VIDEO_ID"

Process a Local Audio File

python enhanced_speech_tool.py -f "path/to/audio/file.mp3"

Command Line Options

• -yt, --youtube: YouTube URL
• -f, --file: Local audio file path
• -u, --url: Direct audio URL
• -c, --config: Path to configuration file
• -v, --voice: Voice to use for synthesis
• --no-disfluencies: Remove disfluencies (um, uh, etc.)
• --simplify: Simplify language for easier understanding
• -o, --output-dir: Output directory

Voice Options

Several high-quality voices are available through Piper TTS:

• en_US-lessac-medium: Clear American English (default)
• en_GB-alba-medium: British English
• en_US-ryan-high: Male American voice
• en_AU-sydney-medium: Australian English

Configuration

You can customize the tool's behavior by creating a configuration file. Example:

{
    "output_dir": "enhanced_audio",
    "whisper_model": "base",
    "device": null,
    "remove_disfluencies": true,
    "simplify_language": false,
    "tts_engine": "piper",
    "voice": "en_US-lessac-medium",
    "maintain_timing": true
}

Project Structure

enhanced_speech_tool/
├── enhanced_speech_tool.py     # Main script
├── setup.sh                    # Setup script
├── config/                     # Configuration files
│   └── default.json            # Default configuration
├── src/                        # Source code
│   ├── audio_extractor.py      # Audio extraction module
│   ├── transcriber.py          # Speech transcription module
│   ├── text_processor.py       # Text processing module
│   ├── speech_synthesizer.py   # Speech synthesis module
│   ├── audio_mixer.py          # Audio mixing module
│   └── config.py               # Configuration module
└── enhanced_audio/             # Output directory

Limitations

• Whisper transcription may not be perfect for all accents or poor audio quality
• Neural TTS voices require downloading models (~200MB per voice)
• Processing long audio files (>30 minutes) can take significant time

Credits

This tool uses the following open-source libraries:

• OpenAI Whisper for transcription
• Piper TTS for speech synthesis
• yt-dlp for YouTube audio extraction
• PyDub for audio processing