Features

• AI-Powered Noise Reduction: Custom U-Net model trained to remove background noise
• Automatic Transcription: Whisper integration for accurate speech-to-text
• Text-to-Speech: Generate synthesized speech from transcripts
• Fast Processing: Optimized pipeline with GPU support
• High Quality: PESQ score of 2.83-2.91
• Flexible: Enhancement-only, transcription-only, or both
• Multiple Output Options: Original enhanced audio or TTS version

We are a Science Fair group dedicated to improving people's lives through technology, and this application is the result of that goal. Our team consists of three students—Aditya, Josh, and Advik—and we all attend Quail Valley Middle School. The objective of this project was to design and develop an AI model capable of clarifying distorted or slurred speech. Such speech issues can arise from technical limitations, environmental interference, speech disorders, or other conditions that affect clarity and intelligibility. Our aim was to create a tool that can help make spoken communication more accessible and easier to understand. Throughout this project, we invested significant time and effort into researching, coding, testing, and training the model ourselves. The development process involved overcoming numerous challenges, refining our approach, and continuously improving accuracy and performance. We are proud of the progress we have made and the results achieved. This project represents months of dedication and collaboration, and we hope that this application proves useful to others. Thank you for taking the time to explore and use our work.

To contact our group to ask questions, give reviews, give suggestions and more, use the following emails.
info@clearspeech.app

More information about this project can be found on the GitHub repository:

Additionally, the model is available on Hugging Face:

Finally, the backend hosting is also on Hugging Face!