Otter.ai Adds 27 Languages, but Automatic Language Switching Is Still on the Roadmap
Otter.ai has announced a transcription feature designed to keep up when speakers switch languages within one meeting. The company is adding support for 27 languages in what it calls its largest language expansion to date. Once the rollout is complete, Otter will support 31 languages in total, according to TechCrunch.
Thirteen of the new languages are available now: Mandarin Chinese, Korean, German, Italian, Cantonese, Portuguese, Russian, Indonesian, Malay, Dutch, Turkish, Vietnamese and Polish. Fourteen more are listed as coming soon: Swedish, Danish, Finnish, Arabic, Hindi, Thai, Hebrew, Norwegian, Greek, Macedonian, Persian, Czech, Hungarian and Romanian.
The switching capability is the more demanding part of the announcement, and it has not shipped. Otter says the upcoming feature will detect language changes automatically and keep transcribing without users adjusting settings. Because it is described as forthcoming, the announcement is a statement of intent rather than a working function. That distinction matters for teams planning to rely on it in multilingual calls.
The new languages run on Otter's own speech models rather than a third-party recognition engine. According to the same report, Otter trained those models on publicly available datasets. Native-speaking volunteers then reviewed transcripts and checked accuracy.
Most transcription products are configured around a single language per session. A meeting that moves between languages has typically required manual reconfiguration or accepted degraded output. Automatic detection at the utterance level, meaning per spoken phrase rather than once per session, shifts that burden onto the model. Code-switching, where a speaker moves between languages mid-sentence, is harder again, and the announcement does not say how the system handles it.
The volunteer review process is worth noting. Public corpora, the large collections of recorded speech used for training, offer broad coverage but often underrepresent regional accents, domain vocabulary and informal speech. Checks by native speakers are one way to find those gaps before release. How far that review scales across 27 languages, and how accuracy compares with other systems, are questions the announcement leaves open.
In my view, the most useful test of this release will be the first few weeks of real multilingual meetings, not the language count. A 31-language list reads well in a changelog. The question for a product manager is whether a sales call that drifts from English into Turkish, or a standup that alternates between Mandarin and English technical terms, produces a transcript someone can search and act on without cleanup.
The broader context here is that transcription now sits beneath meeting summaries, action-item extraction and retrieval systems. Errors at the transcription stage carry downstream, so language coverage becomes a dependency for everything built on top. For organisations whose meetings already cross language boundaries, automatic switching removes a configuration step people often forget, which should mean fewer silently wrong transcripts. For the rest of the market, the announcement is another sign that the language barrier in speech-to-text is narrowing, one model release at a time.
The rollout is staged. Thirteen languages are live, fourteen are announced for later, and the feature that ties them together is still on the roadmap.
