Streamlit demo project for Openai Whisper.
Streamlit App works - Has some issues - Last changed: 2023-03-10
- Dependencies of
whisper
are very heavy (pytorch, transformers, cuda...) and take a long time to install. - Model is not pre-downloaded and very huge, so it takes a long time to download the first time.
- Larger models above the
small
model seems to crash on streamlit cloud, probably due to memory limits. - The largest model that seems to run locally and on streamlit cloud is the
small
model. - Usage of Git-LFS for pre-downloaded model also does not make sense, because at least in the free version of GitHub, you only get 1GB of storage space and 1GB of bandwidth per month.
- The transcribe process takes a long time, probably due to the lack of GPU support.
- The currently used audiorecorder component is not working. Have to check out other alternatives.
- Test app locally with docker
- Test app locally within virtual environment
- Use temporary directory for uploaded file
- Try deployment on streamlit cloud
- Try larger models
- Add custom audio recording component to streamlit app
- Fix and test audiorecorder issue
- Try other audio recording components
- Improve visual layout of the frontend
- Restructure file structure
- Streamlit ❤️
- Openai Whisper ❤️
- Streamlit Custom Components for Audio Recording
- streamlit-audiorecorder 🤷♂️
- audio-recorder-streamlit 🤷♂️
- streamlit-mic-recorder 🤷♂️
- streamlit-webrtc 🤷♂️
- streamlit_audio_recorder 🤔