anime dubbing cli tool Used AI

now you can change the max speed factor ( which affect when tts oversize

I'm happy to tell you that the XTTS here has better tts performance

We have a brand new log

Alpha version here the audio is quite bad, which is not good for a podcast but anime is not. Hopefully i can make that better ( maybe not )

Introducing the stage-based pipeline, which helps you resume when a crash occurs. The new TTS lib added is Edge-TTS, not the best at voice cloning, but that good at clear talking

I decide to use F5-TTS for now ( maybe i will add more ). Some finetuned model of F5 have the CC-BY-NC copyright so careful when you use

after many time refactor we can continue to transcript

By some reason the loss is so high, so I decided to use WhisperX to decrease the loss

I try different approaches. By the way, the first pipeline uses Speaker-Diarization-3.1, but I found that when many people talk at the same time, it poses a significant problem. So I switched to vad with pyannote/speech-separation-ami-1.0 and voice embedding to make that more available. In that case, the new solution will take every object to a depth is 1 (But the loss is so high, hopefully that descrease in next devlog)

That can look for the emotional transcription maybe that