Microsoft Launches In-House AI Model: MAI-Voice — Audio Generated in Just One Second

Microsoft has unveiled its first two self-developed AI models, MAI-Voice-1 and MAI-1-preview. MAI-Voice-1 requires only a single GPU and can generate one minute of audio in just one second. It has already been integrated into “Copilot Daily,” and users can try it out on the Copilot Labs platform, where they can also customize voice and style. MAI-1-preview, designed for users with specific needs, was trained on approximately 15,000 NVIDIA H100 GPUs and features strong instruction-following capabilities.