Microsoft Launches 'MAI-Transcribe-1' Model With Industry-Leading Speech-to-Text Accuracy

Microsoft has just unveiled MAI-Transcribe-1, its latest speech-to-text model that offers highly accurate transcription across real-world audio conditions. The model, developed by the Microsoft AI Superintelligence team, supports 25 languages.

Here’s the list of supported languages: English, French, German, Italian, Spanish, Hindi, Portuguese, Czech, Danish, Finnish, Hungarian, Dutch, Polish, Romanian, Swedish, Japanese, Korean, Chinese, Arabic, Indonesian, Russian, Thai, Turkish, and Vietnamese.

Additionally, the model is optimized for accents, dialects, and noisy environments, making it a versatile tool for developers building voice-enabled applications.

Microsoft said that MAI-Transcribe-1 outperforms competitors such as Scribe v2, Whisper-large-V3, GPT-Transcribe, and Gemini 3.1 Flash on the FLEURS benchmark, achieving best-in-class Word Error Rate across all supported languages.

Its capabilities span live captions, call center transcription, video subtitling, accessibility tools, e-learning, media archiving, and market research, providing text alternatives that enhance usability and compliance. The model can automatically identify languages and is robust in real-world noisy situations, though some advanced features, including real-time transcription, speaker diarization, and context biasing, are slated for future releases.

MAI-Transcribe-1 accepts audio in WAV, MP3, and FLAC formats and is priced at $0.36 per hour of audio processed. Developers can access MAI-Transcribe-1 via Azure Speech, using either the Speech SDK or REST APIs when needed.

As of now, usage requires pointing to East US or West US regions, but Microsoft plans to expand global availability soon. With its strong transcription accuracy and broad language support, MAI-Transcribe-1 looks to power a range of applications from accessibility enhancements to content creation and customer service analytics.