How to enable timecode and speaker diarization?

[bergstar]

Hi! How can I enable or instruct using prompt to enable the timecode output with speaker diarization?

[Benioh]

Hi! You can try prompting it like this:

For speaker diarization only:

Transcribe the audio into text. Each sentence should begin with a speaker label ([S01], [S02], [S03], …), followed by the corresponding spoken content.

For speaker diarization with timecodes:

Transcribe the audio into text. Each sentence should begin with a start timestamp and a speaker label ([S01], [S02], [S03], …), followed by the corresponding spoken content, and end with an end timestamp to clearly indicate the time span of that sentence.

That said, timestamped diarization is currently not one of MOSS-Audio’s strongest settings, so the quality may be only moderate. For this use case, we recommend trying our MOSS-Transcribe-Diarize (MTD) model instead:

http://mosi-ai.com/models/moss-transcribe-diarize

We are also planning to release an open-source version of MOSS-Transcribe-Diarize built on top of MOSS-Audio as the backbone model. Stay tuned!