Full Process Solution for Long Audio Processing
The system will report an error when the audio exceeds 3 minutes:
- hardware solution
Upgrade your graphics card to an RTX3060 or higher model with at least 12GB of video memory to ensure:- CUDA version ≥ 11.8
- PyTorch with cudnn acceleration enabled
- Software adjustments
Modify key parameters:- Find the max_seq_len parameter in models.py
- Recommended Value:
- 5 minutes of audio: set to 6144
- 10 minutes of audio: 12288
- Synchronized modification of the corresponding parameter of llama3_2_100M()
- alternative
Split long audio using ffmpeg:ffmpeg -i long.mp3 -f segment -segment_time 180 -c copy out%03d.mp3
This answer comes from the articleCSM Voice Cloning: Fast Voice Cloning with the CSM-1BThe
May not be reproduced without permission:AI productivity tools " How to overcome the tensor dimension error in CSM Voice Cloning when processing long audio?
