I developed a web application that automatically transcribes audio files and identifies different speakers in conversations using OpenAI’s Whisper model and speaker diarization technologies, creating a powerful tool for transcription needs.
Text to Audio Transcription with Speaker Diarization
This project combines modern web technologies with state-of-the-art AI models to create an easy-to-use tool for transcribing speech with speaker identification. Key features include:
- Modern User Interface:
- Clean, intuitive web interface with drag-and-drop file uploads
- Real-time progress tracking during transcription processing
- Responsive design that works on various devices
- Support for 18 languages in the user interface
- Powerful Transcription Engine:
- Integration with OpenAI’s Whisper speech recognition models
- Multiple model sizes (tiny, base, small, medium, large) to balance speed and accuracy
- Speaker diarization to identify and separate different speakers in the conversation
- Support for various audio formats (MP3, WAV, OGG, FLAC, M4A)
- Flexible Output Options:
- Download transcriptions in multiple formats (TXT, JSON, SRT, VTT)
- Timestamped transcripts with speaker identification
- Clean formatting for easy reading and post-processing
This tool significantly improves productivity for anyone working with audio transcription - from journalists and researchers to content creators and accessibility professionals.
The application is built using Flask for the backend, with a modern JavaScript frontend. It incorporates several advanced technologies including PyTorch, OpenAI Whisper, and Pyannote Audio for speaker diarization.
