This project is a tool that was developed as part of a Master's thesis in cooperation with the University Clinic of Psychoanalysis and Psychotherapy of Vienna. It is based on the Whisper automatic speech recogniton system and is embedded into a Streamlit Web App.
- Streamlit UI: The tool includes a user-friendly interface that allows you to upload multiple audio files and get a nicely formated transcript.
- Pause detection: The tool can detect pauses in the audio.
- Confidence scores: The tool can color the words according to its probability and display the average score.
- Translation to english
- Speaker detection: not available
- Whisper is used locally as well as offline (no internet connection needed)
- Nothing is being uploaded to the cloud
- Therefore safe clinical use
To use this tool, you will need to install the required dependencies and run the Streamlit app. You can do this by following these steps:
- Clone the repository: git clone https://github.com/jojojaeger/whisper-streamlit
- Install prerequisites: Python, Pip, Git, PyTorch (pip install torch torchvision torchaudio)
- Install dependencies: pip install -r requirements.txt
- Run the Streamlit app: streamlit run Transcribe.py (you can also launch it from a desktop shortcut following these instructions: https://discuss.streamlit.io/t/launching-streamlit-webapp-from-desktop-shortcut/26297)
- Upload one or multiple audio files
- Select a model (large for the best result) and set additional parameters
- Download the resulting transcript (also saved to local transcripts-folder)
- Error: Request failed with status code 403 -> run the app with "--server.enableWebsocketCompression=false"
If you have any questions or feedback about this project, please feel free to contact us by email at johanna.jaeger89@icloud.com.
This project includes code from multiple different sources, each licensed under the MIT License:
- [Source A] (https://github.com/openai/whisper)
- [Source A] (https://github.com/pyannote/pyannote-audio)
- [Source C] (https://github.com/hayabhay/whisper-ui)
See the LICENSE file for the full text of the licenses.