Microsoft has introduced MAI-Voice-2, a new text-to-speech model designed to generate speech in 15 languages. The company is positioning the system as part of its broader Microsoft AI efforts, with access available through its model exploration and playground tools.

The launch highlights Microsoft’s continued push into generative AI, this time focused on spoken audio rather than text or images. MAI-Voice-2 is presented as a multilingual speech generation model, indicating that it can convert text into spoken output across a wide set of languages. Microsoft did not provide detailed technical specifications in the source material, but the release makes clear that the model is meant for experimentation and testing.

Microsoft is offering the model through its Foundry model discovery experience and through MAI Playground, where users can try experimental AI systems. The company notes that MAI Playground is a limited preview, signaling that the voice model is not yet a fully mature consumer product. As with other early-stage AI tools, Microsoft cautions that the system can make mistakes.

The sample content associated with the launch shows the model being used to read a short scene in a calm, descriptive tone. The example suggests that Microsoft is emphasizing voice quality and natural narration, in addition to language coverage. That focus may make the model relevant for uses such as narration, accessibility tools, customer support, or other speech-based applications, though Microsoft did not spell out specific product plans in the source.

Multilingual speech tools have become an important area of competition among major AI developers, especially as companies look for ways to make their systems more useful across global markets. A model that can produce speech in multiple languages could help broaden adoption among developers and enterprise customers, particularly if it can deliver consistent voice output and flexible deployment options.

Microsoft has been steadily expanding its portfolio of proprietary AI models under the MAI label. The release of MAI-Voice-2 suggests the company is continuing that strategy by building models tailored to specific modalities and use cases. Rather than focusing only on general-purpose chat, the company is also adding tools for audio generation and other specialized tasks.

The preview status of the model means Microsoft is still collecting feedback and testing how the system performs. That approach gives the company room to refine the technology before any broader release. It also reflects a common pattern in AI development, where new capabilities are introduced first as experimental offerings before they are integrated into more polished products.

For now, MAI-Voice-2 appears to be aimed at developers and AI enthusiasts who want to test multilingual speech generation capabilities. Microsoft has not publicly detailed pricing, availability limits, or a release timeline for a broader rollout in the material provided. Still, the launch adds another piece to the company’s growing AI model lineup and underscores how quickly speech generation is becoming a core part of the generative AI market.