Anyone get the looped text in the output of whisper?
Currently we meet around 5-10% of recording files get the looped output in Chinese speech recognition.
2 Likes
Same here for Japanese transcription as followings.
Iβm feeling this kind of loop happens very often. In my case, about 20% of all transcriptions.
... γ‘γͺγΏγ«θΏδΊζ₯γγγ£γγγ© γͺγζ₯γͺγγ£γγ γγγ¦γγγγγγΌγγ£γ¦δ½ γͺγγζε₯γͺγγ γ ι Όγγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ γεγ ...
Related message in discord:
Hello. Have you continued to work with Whisper? And did you have any luck in solving this problem? I am facing the similar issue in Serbian, Croatian and some other smaller languages.