feldberlin / timething

Timething is a library for aligning text transcripts with their audio recordings.
MIT License
101 stars 7 forks source link

Works on a podcast #5

Closed purzelrakete closed 1 year ago

purzelrakete commented 2 years ago

What

Can run on a 30 minute Podcast in a single mp3.

Why

Currently only tested on pre-cut batches. Longer content can lead to OOMs, quadratic time cleaned alignment issues, or poor audio alignment performance. A real world transcript may also fail to align due to issues like intra-transcript inserts, e.g. speaker names.

Acceptance Criteria