transcribe task output translate result

[wylqq312715289]

my transcribe code is

model.transcribe(test_pcm, language="en", beam_size=4, word_timestamps=True, temperature=0)

the audio contain chinese voice and speak chinese, but the result was translated in english!

[guillaumekln]

language="en"

The language parameter must be the language of the audio. So you should set zh here.

[wylqq312715289]

language="en"

The language parameter must be the language of the audio. So you should set zh here.

but the audio contains english and chinese too. but the most time is english, only 5% contents is chinese.

[guillaumekln]

Whisper is not designed to work with multiple languages in the same audio.

See openai/whisper#49 for possible solutions.