Details
- Microsoft AI announced MAI-Voice-2, its latest text-to-speech model focused on production-grade voice quality across languages.
- The model supports 15 languages with an expressive range designed to maintain naturalness and speaker identity across all supported locales.
- Internal side-by-side evaluations show MAI-Voice-2 is preferred over MAI-Voice-1 in 72% of comparisons, indicating a clear quality improvement.
- Microsoft highlights gains in fidelity, expressiveness, and speaker consistency, describing MAI-Voice-2 as a step change for production voice use cases.
- MAI-Voice-2 can code-switch mid-sentence for Hindi-English and Spanish-English, preserving natural rhythm, tone, and speaker identity in mixed-language markets.
- Developers can start building with MAI-Voice-2 via Microsoft Foundry and OpenRouter, with pricing set at $22 per 1 million characters.
- According to the product blog, MAI-Voice-2 is designed for applications such as assistants, customer support, audiobooks, games, accessibility, and creator workflows where natural, controllable voice is critical.
- The blog also notes that MAI-Voice-2 brings more granular control over expressiveness and emotional range compared with its predecessor, supporting more nuanced and branded voice experiences.
Impact
MAI-Voice-2 strengthens Microsoft’s position in the fast-evolving speech generation market by upgrading quality while keeping pricing aligned with MAI-Voice-1. The addition of robust multilingual support and code-switching directly targets global, consumer-facing and enterprise applications, putting competitive pressure on other large providers of neural TTS to match mixed-language and expressive control capabilities.
