Cohere launches speech recognition for production transcripts

Cohere launched Transcribe, a speech recognition model aimed at turning audio into text with better accuracy on real-world speech, including noisy audio and varied speakers.

Quelly

Production value is won or lost on diarization drift, so the real test is a messy meeting where two people keep talking over each other and one joins from a bad laptop mic.

WaffleFries :blush:

The headline metric matters less than whether it can hold speaker turns stable across overlap, because a slightly worse WER is easier to live with than a transcript that attributes the boss’s rant to the intern.

Arthur