Completed
0:00 Whisper preparation and fine-tuning with Unsloth
Class Central Classrooms beta
YouTube videos curated by Class Central.
Classroom Contents
Whisper Data Preparation and Fine-Tuning with Unsloth
Automatically move to the next video in the Classroom when playback concludes
- 1 0:00 Whisper preparation and fine-tuning with Unsloth
- 2 0:40 Resources: Trelis.com/ADVANCED-audio
- 3 1:23 One-click GPU and Jupyter Notebook Setup
- 4 3:37 Whisper vs Voxtral vs Kyutai
- 5 4:48 Installation of Unsloth and Whisper Timestamped
- 6 7:52 Using Whisper Large versus Turbo
- 7 8:53 Video Overview / Layout - How to prepare data and train
- 8 11:33 Audio recording and transcription with whisper timestamped
- 9 13:06 Whisper vs Whisper-Timestamped and the motivation for word timestamps
- 10 15:34 Creating text/audio segments using word-timestamped transcripts
- 11 18:46 Segment time-stamps using whisper not easy to then chunk to less than 30s!!!
- 12 19:50 Word time-stamps with Whisper Timestamped
- 13 20:56 Automated vs manual transcript cleanup techniques
- 14 28:48 Dataset creation from audio and text segments
- 15 20:53 Fine-tuning with Unsloth
- 16 33:26 Word Error Rate - Teacher Force versus predict_with_generate
- 17 36:11 Training hyperparameters and losses / results
- 18 37:33 Evaluating base and fine-tuned model performance
- 19 39:15 Merging, pushing to hub and preparing for inference see also https://www.youtube.com/watch?v=qXtPPgujufI
- 20 40:21 Conclusion
