🎙️ VoiceGenMeeting

VoiceGenMeeting is a command-line tool that generates synthetic meeting audio from a simple text-based transcript, assigning a unique voice to each speaker. It's ideal for testing transcription, meeting analysis, or speech recognition tools.

TTS Engine Versions

This project offers multiple implementations using different text-to-speech (TTS) engines. Choose the version that best suits your specific requirements:

1. Coqui-TTS Branch lib/coqui-tts

• Recommended for: Multilingual projects, especially non-English content
• Strengths:
  • Exceptional phonetic accuracy across languages
  • Robust language support
• Limitations: Slightly monotonous intonation

2. Kokoro Branch lib/kokoro

• Recommended for: English-language projects
• Strengths:
  • Nuanced intonation
  • Expressive English speech synthesis
• Limitations: Poor handling of non-English phonetics

3. OuteTTS Branch lib/outetts

• Based on: Open-source TTS engine using llama.cpp and Transformers
• Good for: Experimental and lightweight use cases

Choosing the Right Version

• Multilingual Project: Use Coqui-TTS branch
• English-Only Project: Use Kokoro branch
• Experimental or Resource-Constrained Setup: Use OuteTTS branch

We continuously evaluate and improve these implementations. Future versions may integrate the best features from each engine.

Example Comparisons

Refer to the example.wav and example_fr.wav files in each branch to hear the differences in voice synthesis quality.

Features

• Generates a .wav file from a .txt transcript formatted like:

  Marc : Hello, shall we begin?
  Julie : Yes, I'm ready.
  

• Automatically assigns a voice to each speaker
• Creates custom voice profiles from short audio files (marc.mp3, julie.wav, etc.)
• Saves and reuses speaker profiles from the data/profiles folder
• Converts and truncates source audio files to a maximum of 15 seconds
• Fully offline — no cloud API dependency thanks to OuteTTS

Project Structure

VoiceGenMeeting/
├── conversation_tts.py       # Main CLI script
├── data/
│   ├── speakers/             # Raw speaker audio files (e.g. marc.mp3)
│   └── profiles/             # Generated speaker profiles (.json)
├── example*.txt              # Example transcript with multiple speakers
└── example*.wav              # Example result of generated audio output

License

MIT