Microsoft has released three specialized artificial intelligence (AI) models, focusing on image generation, voice generation, and speech-to-text transcription.
The models, MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2, are also said to focus on fast generation and competitive pricing.
All of them are currently available via Microsoft Foundry and the MAI Playground. The biggest highlight is the MAI-Transcribe-1, which the company claims delivers state-of-the-art (SOTA) speech-to-text transcription across the 25 most used languages.
The claims are based on Microsoft’s internal testing using the FLEURS benchmark, where it is reported to outperform both Gemini 3.1 Flash and GPT-Transcribe in terms of error rate. Additionally, the company says Foundry users will find it to be the “best price-performance of any large cloud provider.”
The MAI-Voice-1, the LLM is said to generate “natural, realistic speech, rich with nuance, emotional range, and expression.” The model is also said to deliver consistent speech and voice identity during long-form content generation. Inside Foundry, the model will also allow users to create a custom voice with a few seconds of audio.
According to Microsoft, this process is safe and secure. It is said to generate 60 seconds of audio in a single second. Notably, the AI model will also power Copilot Audio Expressions and Copilot Podcasts.
Finally, the MAI-Image-2 model builds on the capabilities of its predecessor and is said to deliver improved output quality at a faster speed.
Microsoft revealed that the model was created in collaboration with photographers, designers, and visual storytellers, and it focuses on natural lighting, accurate textures, and clear in-image text. WPP is among the first enterprise partners to have adopted the AI model, as per the statement.
