# MAI-Transcribe-1.5 expands language coverage and adds keyword biasing > Microsoft’s transcription update covers 43 languages and introduces domain vocabulary hints, while streaming and diarization remain roadmap items. Canonical URL: https://www.devobs.io/news/news-mai-transcribe-1-5-launch/ By: Ines Costa Published: 2026-09-06T11:58:54.629Z Updated: 2026-09-06T11:58:54.629Z Event date: 2026-06-02 Section: AI Microsoft launched MAI-Transcribe-1.5 on June 2 with expanded coverage across 43 languages and a keyword-biasing feature for specialized vocabulary. The [release announcement](https://microsoft.ai/news/mai-transcribe-1-5more-accurate-context-aware-and-built-for-production/) said the model was available in Foundry and being integrated into Microsoft products. ## Vocabulary becomes an explicit input The update allows callers to provide a list of domain-specific words. Microsoft described that list as contextual guidance rather than a rule that forces every similar sound into a supplied term. The intended examples include names, product terminology and internal abbreviations. The post listed speaker diarization and a native streaming API under future work. Those capabilities should therefore not be treated as features established by this launch. ## Test hints for both help and harm Our suggested evaluation is a paired test: transcribe each recording with and without the same vocabulary list. Track whether the intended names improve, but also inspect ordinary words that sound similar to the hints. A dictionary that helps one department may be inappropriate for another. Keep the vocabulary list small enough to explain and version it alongside the recording’s processing metadata. A later investigation should be able to distinguish a model upgrade from a changed customer glossary. Include multilingual examples where a product name is spoken inside another language rather than assuming a monolingual test covers that case. ## Keep the processing boundary clear A batch transcription service and a live conversation interface need different acceptance tests. Teams should identify whether their application waits for an entire recording or needs useful text while a person is still speaking. For this release, the most concrete integration question is whether contextual vocabulary improves the recordings the team already processes. That can be tested independently of the vendor’s roadmap and without redesigning the whole audio workflow. ## Source references -