Piloting the Soniox 2nd time with their new(w15) release

[beastoin]

Objective: add support the multi-language to the omi system

Key results:

• adapt to the new api spec
• test the hallucination which is the cause of long transcripts leading to high OpenAI billing.

--

BBernard — 16/04/2025, 03:25
Important ---- new multilingual Speech-to-Text was just released in production. Both Real Time and Async models were updated and so is API and docs. Most changes relates to real time API - please check the documentation (https://soniox.com/docs/speech-to-text/get-started) and update with new APIs. We are here to support you with any questions you might have. For maximum accuracy use language hints (several could be selected simultaneously), speaker separation in real time model is now greatly improved, wer accuracy was already very high and is now even better, both in async and real time and across all languages.