maan213/speech-fluency

-

🧠 Disordered to Fluent Speech Converter

This project uses machine learning models to convert disordered speech into fluent, intelligible speech. It utilizes OpenAI's Whisper for automatic speech recognition and Coqui's Tacotron2-DDC for text-to-speech synthesis. The system also computes the Word Error Rate (WER) to assess how well the original speech is preserved in meaning.

Features

• —Transcribes disordered speech using Whisper.
• —Synthesizes fluent speech using Tacotron2-DDC.
• —Computes Word Error Rate (WER).
• —User-friendly interface built with Gradio.

Models Used

• —Whisper (OpenAI) – for speech recognition.
• —Tacotron2-DDC (Coqui TTS) – for speech synthesis.

How to Use

1. 1.Upload a speech file with disordered or impaired speech.
2. 2.View transcriptions and synthesized fluent speech.
3. 3.Check the WER score to evaluate conversion accuracy.

Requirements

All dependencies are listed in requirements.txt.

License

For research and educational purposes.