New Azure AI Speech feature enables speech-to-speech transformation without altering tone or emotion

Microsoft has unveiled a new voice conversion feature in Azure AI Speech that allows users to transform a recorded voice into a different AI-generated voice — while preserving the original rhythm, tone, and emotional expression. The feature is currently in preview and marks a shift from traditional text-to-speech models toward more dynamic speech-to-speech processing.

Unlike previous approaches that rely on text input, this technology repurposes the intonation and expressiveness of the original audio, changing only the speaker identity. It is initially available in select regions, including Eastern US, Western Europe, and Southeast Asia, and supports 28 EN-US voices also used in Microsoft’s existing speech services.

Aiming for consistency in multilingual dubbing

A key application of the feature is multilingual dubbing. Microsoft notes that localized audio often varies in style and clarity across languages. With voice conversion, all dubbed versions of content can share a single, consistent target voice — offering a more uniform experience across linguistic markets.

In internal testing, Microsoft reports that the system outperformed a leading competitor in Mandarin, producing more natural-sounding and clearer results. In English, the performance was said to be on par.

Customization and implementation options

The feature is also being integrated into Microsoft’s Custom Voice platform, now available in private preview. This extension allows companies to apply voice conversion to their own branded synthetic voices using only limited training data. The approach aims to simplify dynamic voice personalization while maintaining emotional fidelity and consistency.

Developers and content teams can explore implementation details and technical guidance through Microsoft’s Azure AI blog.