Seeking Advice on Audio Segmentation & Fine-Tuning for Improved Korean Pronunciation

[jikerbug]

Hello! Thank you for all your amazing work!

I tried using the JP-KR model, but possibly due to the mix of both languages, I haven’t noticed a clear improvement in Korean pronunciation compared to the original model. Therefore, I’m preparing a LoRA-based fine-tuning process specifically to enhance Korean pronunciation. Right now, I’m slicing audio into fixed 30-second segments, but this often ends up cutting lyrics in the middle. Would you recommend aligning segments precisely with the lyrics instead, or does a static-length approach still work well in your experience? Any insights you can share would be greatly appreciated.

Once again, thank you for your incredible contributions!