AI BriefingCohere STT Now Supports Speaker Diarization
Key point
A fine-tuned version of Cohere's open-source speech recognition model Transcribe has been released, adding speaker diarization and timestamp features.
Details
A fine-tuned version of Cohere Transcribe, regarded as the top-performing open-source speech-to-text model, has been released for free on Hugging Face, adding speaker diarization and timestamp capabilities.
The existing Cohere Transcribe did not support speaker identification or time information output, even though the tokenizer included the relevant tokens. This fine-tuned model follows a standard timestamp format and produces output such as the following:
<|spltoken0|><|t:0.0|> Welcome back. <|t:1.5|><|spltoken1|><|t:1.5|> Thanks. <|t:2.4|>
Key performance metrics:
- Timestamp accuracy: within 0.097 seconds on average, within 0.006 seconds for 90% of cases
- Supports distinguishing up to 4 speakers per 30-second segment
- Using the
diarize_long.pyscript, up to 32 speakers can be identified
The model is freely available on Hugging Face.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Qwen3.5-based Small Decision Model 'Kev' Released with TypeSafe API Compatibility and Performance MetricsJevper Released: A Jev Classification Wrapper for OpenAI-Compatible Models with Automatic Logprobs Detection