Speech-to-text built for conversations with more than one voice with highest accuracy than any in the industry. Lowest diarization error on several benchmarks. 45 languages, up to 25 speakers. Add-ons: subtitles, PII redaction, summaries, speaker names, translation. Better than Pyannote, Deepgram and Assembly AI

Hey Product Hunt, we're the team behind Maybloc.

Most transcription tools handle one voice well and fall apart when people talk over each other. We built Maybloc for that case: every speaker kept apart, overlapping speech kept instead of dropped, every word timed.

Under it is Calico, our speech model, which builds on open research including DiCoW from BUT. On the public multi-speaker benchmarks we ran (NOTSOFAR, AMI, DiPCo) it has the lowest siarization (who-said-what) error of pyannote, AssemblyAI and Deepgram, and over 40% fewer errors than AssemblyAI and pyannote on NOTSOFAR's real office meetings. Tables and method: maybloc.com/#benchmarks

Use it in the dashboard or through a REST API. 45 languages, up to 25 speakers, files up to 3 hours, $0.006 per minute billed by the second. Add-ons: subtitles, PII redaction, summaries, speaker names, translation.

Try it on your messiest recording and tell us where it breaks. Happy testing!