Speechmatics have added a new language to the speech recognition (ASR) roster: Persian. Significantly better performance than other providers and pushes AI to understand more than American English voices.
Also 25% - in our testing we found Whisper's medium.en model to give similar WERs to Whisper's large_v2 model. On some testsets though, especially Switchboard and AMI-IHM, Ursa achieves >40% lower relative WER compared to Whisper's medium.en model
We are excited to introduce Ursa, the world’s most accurate speech-to-text system. Ursa delivers unprecedented performance across a diverse range of voices. We observe relative accuracy gains of 22% and 25%* versus Microsoft and OpenAI’s Whisper respectively.
