Microsoft announced on October 26, 2026 that Azure LLM Speech is adding 25 languages, including 11 Indic languages, to its transcription model. The expansion gives creators making captions and businesses processing customer recordings more language options within this service.
The named Indic additions are Assamese, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu and Urdu. Microsoft’s LLM Speech language table lists Indian locales for each; support for a language name should not be read as support for every regional variant.
This is an expansion of a particular model, not Azure’s first Indic transcription capability. Separate Azure release notes date another multilingual Indic speech model’s availability to October 5.
Check the language of the recording—and the subtitles
Transcription writes down speech; translation changes it into another language. Tamil recordings are now listed for this model’s transcription, but Tamil is not among its documented translation outputs. English is an output option; none of the 11 newly named Indic languages is.
For creators, the practical implication is to check two things separately: the recording’s spoken language and the intended translated subtitle language. Support for transcribing Tamil does not mean the same model can translate an English recording into Tamil subtitles.
For recordings that mix supported languages, Microsoft’s usage documentation says to leave the optional language setting unset so the model detects them automatically. Specifying a known language can help with single-language audio. Language detection is an existing feature, not something introduced by this expansion.
Automatic rollout still depends on the service you use
Microsoft says the update is available through Fast Transcription and will roll out automatically without customer action. But LLM Speech uses the service’s enhanced mode: an application using default transcription does not become an LLM Speech application simply because the model has changed.
The overview still identifies the current model as 07 2026, while the announcement names 09-2026. The version labels differ, so Microsoft’s rollout statement should not be treated as confirmation that an individual account has already received the update.
The Foundry playground offers a no-code route, requiring an Azure subscription and Foundry project. For API use, a Speech resource must be in an eligible region. The LLM Speech region list currently includes Central India, East US, North Europe, Southeast Asia, West US and West US 2.
Microsoft also reports up to twice-as-fast processing, particularly for many 5–10-second clips processed at once. That vendor result is not a promise of halved turnaround for every recording, nor evidence of equal accuracy across the added languages.
The pricing page says LLM Speech shares Fast Transcription’s price and bills by audio duration. The public table shows placeholders rather than a usable regional quote; check the price for the intended region and currency before budgeting. The billing implication: at an unchanged rate, the same hour of audio costs the same even if processing finishes sooner.
For businesses considering what happens after transcription, our Granite Switch coverage examines using voice-message transcripts for summaries and request sorting. Its default speech component is English-only, so it is not an equivalent multilingual option.