Microsoft's MAI-Transcribe-2-Streaming can now be used by LiveKit Agents through an official Microsoft AI speech-to-text plugin. The integration gives developers a documented path to send live audio from a LiveKit voice agent to Microsoft's realtime transcription endpoint and receive continuously updated text. For teams building support, phone, or voice-assistant workflows, the development expands the choice of transcription providers, but the underlying Microsoft service remains in public preview and is not recommended for production workloads.
The core capability is real-time speech-to-text. MAI-Transcribe-2-Streaming is designed to process audio as it arrives, producing incremental transcript updates rather than waiting for a complete recording. Microsoft documents support for roughly 60 languages, along with access through its Realtime API over WebSocket or the Azure Speech SDK. LiveKit's official Microsoft AI STT plugin documentation shows how LiveKit Agents can use that realtime endpoint directly.
How the LiveKit and MAI integration works
LiveKit Agents is a framework for building real-time voice and multimodal agents. Its Microsoft AI STT plugin acts as the connection between an agent session and MAI's realtime transcription service. The plugin streams audio to the MAI endpoint, allowing the agent workflow to consume transcript updates while a caller is speaking.
The documented setup requires three provider-specific values:
- A MAI resource URL for the realtime transcription endpoint.
- An API key used for authentication.
- The model deployment name for the MAI transcription model.
LiveKit's documentation also shows the plugin being used in an AgentSession. That matters because transcription is not a separate, manual export step. It can become part of a live agent pipeline where spoken input is converted to text for the rest of the application to handle.
| Component | Role in the workflow | What the verified documentation confirms |
|---|---|---|
| MAI-Transcribe-2-Streaming | Speech-to-text service | Real-time, low-latency transcription with continuous updates and multilingual support |
| MAI Realtime API | Connection to the transcription service | Realtime API access over WebSocket is supported |
| LiveKit Microsoft AI STT plugin | LiveKit Agents integration layer | Streams audio to the MAI realtime transcription endpoint using configured credentials and deployment details |
What this means for voice support workflows
The practical value is faster access to spoken input as text inside a LiveKit-based agent. A business building a voice support experience could use the transcript as input to its own agent logic, knowledge retrieval, note-taking process, or handoff workflow. The integration itself confirms transcription connectivity. It does not, by itself, establish that calls are automatically routed, resolved, summarized, or less expensive.
That distinction is important for planning. Automated routing would require separate rules or agent logic that acts on the transcript. Cost savings would depend on a company's existing support process, call volumes, staffing, and the pricing of the services used. Neither Microsoft nor LiveKit's cited documentation provides a fixed per-hour price for this MAI transcription integration.
For developers, the supported plugin reduces the work of implementing the provider connection from scratch. It also places MAI alongside other speech-to-text options that can be used in a LiveKit Agents architecture. MAI models have also been noted in communications identifying LiveKit as an ecosystem platform for MAI models, while the plugin documentation provides the more concrete implementation path for streaming transcription.
Availability, pricing, and deployment considerations
MAI-Transcribe-2-Streaming is in public preview. Microsoft states that it is not recommended for production workloads. That status should shape expectations: teams can evaluate the integration and build prototypes, but should not assume preview behavior, availability, or commercial terms are equivalent to a generally available production service.
Pricing requires additional checking before a deployment decision. Microsoft directs users to Foundry model pricing guidance, rather than publishing a fixed per-hour transcription price in the available documentation. A realistic evaluation should therefore confirm the applicable model pricing, test the service with representative audio, and account for the costs of the wider voice workflow, including any LiveKit infrastructure and downstream agent services.
Teams should also validate the operational pieces that the plugin requires. The documented configuration depends on a resource URL, API key, and deployment name. Since the service streams live audio to a provider endpoint, implementation should include deliberate handling of credentials and a clear understanding of where audio is processed in the chosen architecture.
For businesses exploring voice automation, the important near-term question is not whether transcription alone can replace a support process. It is whether reliable live transcripts make a narrowly defined workflow more useful, such as surfacing information to an agent or capturing conversational input for a follow-up process. The preview designation makes measured testing more appropriate than a broad production rollout.
If you are assessing where live transcription fits into a customer or internal workflow, Scalevise can help translate the technical option into a practical implementation plan, identify the process steps that should remain human-led, and connect AI services to the tools your team already uses. Our AI workflow automation service focuses on reducing manual work without forcing an oversized project. Discuss an AI automation project with Scalevise.
Frequently Asked Questions
What is MAI-Transcribe-2-Streaming?
MAI-Transcribe-2-Streaming is Microsoft's real-time speech-to-text model. It is designed to provide incremental transcript updates as audio arrives and supports roughly 60 languages, according to Microsoft's documentation.
Can LiveKit Agents use Microsoft MAI transcription?
Yes. LiveKit documents a Microsoft AI STT plugin that streams audio from LiveKit Agents to the MAI realtime transcription endpoint.
What is required to configure the LiveKit MAI STT plugin?
The documented configuration requires a MAI resource URL, an API key, and the model deployment name. The plugin can then be used in a LiveKit AgentSession.
Is MAI-Transcribe-2-Streaming ready for production support calls?
Microsoft lists the service as public preview and says it is not recommended for production workloads. Teams should evaluate it accordingly before using it in a live production workflow.
Is there a fixed price for MAI transcription in LiveKit?
The available documentation does not provide a fixed per-hour rate for this integration. Microsoft points users to Foundry model pricing guidance, so pricing should be confirmed for the intended deployment.
Conclusion
The MAI and LiveKit integration provides a documented route for bringing Microsoft's live transcription into LiveKit Agents. It is a useful option for evaluating voice-agent workflows that need continuously updated text, but its public-preview status and unspecified fixed integration pricing mean businesses should validate performance, costs, and workflow fit before treating it as a production foundation.
Top comments (0)