SPECTRA - Speech Classification & Transcription Analysis

SPECTRA is a speech analysis tool that can determine whether speech in an audio file is read from a prepared script or spoken spontaneously. It analyzes various speech features including word length, pauses, and speech rate to make this determination.

Features

• Analyze audio files to classify speech as read or spontaneous
• Calculate reading probability as a percentage
• Generate human-readable explanations of analysis results
• Support for multiple audio formats (WAV, MP3, M4A, WebM)
• RESTful API for integration with other applications

Architecture

The project follows a clean architecture pattern:

• src/core/ - Core application logic and DTOs
• src/use_cases/ - Business logic implementation
• tests/ - Unit tests

How It Works

SPECTRA uses a novel approach to classify read and spontaneous speech based on three key features:

1. Active average word length: The average length of words in the speech
2. Inactive alphabets per second: The frequency of pauses
3. Words per second: The overall speech rate

These features are combined to compute a readability score that indicates how likely the speech is to be read from a script.

Dependencies

• Python 3.8+
• FastAPI
• Parselmouth
• Faster-Whisper
• NumPy
• LangChain
• Ollama (for explanation generation)
• FFmpeg (for audio conversion)

Installation

1. Clone the repository:

   git clone https://github.com/mechamogeo/spectra.git
   cd spectra
   

2. Install dependencies:

   pip install fastapi uvicorn parselmouth faster-whisper numpy langchain-ollama
   

3. Install FFmpeg for audio conversion:

   • On Ubuntu: sudo apt-get install ffmpeg
   • On macOS: brew install ffmpeg
   • On Windows: Download from ffmpeg.org

4. Install Ollama for LLM explanations: Follow the instructions at ollama.ai

5. Pull the required model:

   ollama pull granite3.2:2b
   

Running the API

1. Start the API server:

   python -m src.core.app
   

2. The API will be available at http://localhost:8000

3. Access the API documentation at http://localhost:8000/docs

API Usage

Analyze an audio file

GET /analyze?file_name=your_audio_file.wav

Make sure your audio file is placed in the resources/audios/ directory.

Running Tests

To run the test suite:

pytest tests/

References

@misc{kopparapu2023novelschemeclassifyread,
      title={A Novel Scheme to classify Read and Spontaneous Speech},
      author={Sunil Kumar Kopparapu},
      year={2023},
      eprint={2306.08012},
      archivePrefix={arXiv},
      primaryClass={cs.SD},
      url={https://arxiv.org/abs/2306.08012},
}

License

This project is licensed under the MIT License - see the LICENSE file for details.