fix (mistral-ai): add flexibility for timestamps

[tinalenguyen]

timestamp_granularities and language can't be used in tandem. to use one, the other must be set to None

[chenghao-mou]

Thanks for spotting that!

I tested it locally, and it returned timestamps. However, I don't think we can turn the flag on because it is still non-streaming, and therefore only relative to each VAD spans.

STT event: SpeechEvent(type=<SpeechEventType.FINAL_TRANSCRIPT: 'final_transcript'>, request_id='', alternatives=[SpeechData(language=None,        
                                           text=' Can you tell me a different story?', start_time=0.3, end_time=1.9, confidence=0.0, speaker_id=None, is_primary_speaker=None, words=[' Can  
                                           you tell me a different story?'])], recognition_usage=None) 
...
STT event: SpeechEvent(type=<SpeechEventType.FINAL_TRANSCRIPT: 'final_transcript'>, request_id='', alternatives=[SpeechData(language=None,        
                                           text=' You know.', start_time=0.3, end_time=0.7, confidence=0.0, speaker_id=None, is_primary_speaker=None, words=[' You know.'])],                
                                           recognition_usage=None) 

[tinalenguyen]

@chenghao-mou Ah okay, thanks for the clarification! It's always the case that aligned_transcript is set if streaming is True, right?

[ChenghaoMou]

@chenghao-mou Ah okay, thanks for the clarification! It's always the case that aligned_transcript is set if streaming is True, right?

Yes. If it supports streaming and can return word or chunk timestamps, otherwise it should always be False.