benearnthof/podcasty

podcasty's Issues

Preserve Token Uncertainty from Whisper output

For noisy audio (songs etc.) the transcripts may contain errors.

Users should be informed about possible mistakes.

• Preserve Token Uncertainty info from whisper output
• Add functionality to visualize uncertainty in text.

Add other platforms

Currently only youtube is supported, users should be able to transcribe from the most popular podcasting platforms.

Current Scope:

• Spotify (https://github.com/spotDL/spotify-downloader)
• ApplePodcast (not sure if possible yet)
• Soundcloud

Add translation functionality

After the transcription, users should be able to translate the transcript into the most popular languages.
Current Goals:

• English
• German
• French
• Spanish

Add summarization functionality

Users should be able to automatically abridge / summarize transcripts. Hour long podcast conversations can be all over the place so ideally the summarization should output both a high level summary (5-10 sentences) and low level summary for each part.

Possible Approaches:

• Basic Seq2Seq Model (spacy etc.)
• BERT (Nice for local inference)
• ChatGPT (interface with API)

Add bulk functionalities

Currently urls have to be passed one by one. Users should be able to pass in a list of urls or a full playlist and process the contents of the list in bulk.

Current scope:

• Bulk processing of url lists in the form of a .txt
• Bulk processing of playlists by their url

Youtube playlists should be trivial, Spotify playlists should work as long as they are public