In auto mode, the language tag tends to output Chinese

#1
by funnyice - opened

In auto mode, the language tag tends to output <chinese>, although the transcription result is correct.

Language: Auto (tag='')
Text channel: <chinese> um, and then I'll be coming to you ...
FINAL_TEXT: um, and then I'll be coming to you ...

Yes, we suspect this may be because we have assigned the tag to code-switching data. Regardless of the tag used, however, the transcription results under auto mode are already sufficiently accurate. In fact, the Language Tag is designed to offer an option: when you confirm the language in the audio, it provides stronger conditioning for transcription, leading to higher accuracy.

Sign up or log in to comment