MAI-Voice-2.1

Microsoft AI releases new transcription and text-to-speech models

Microsoft AI released MAI-Transcribe-2-Streaming, a real-time transcription model it says ranks first for accuracy on Artificial Analysis, supports 60 languages, returns first partial results in about 100 milliseconds, and costs $0.54 per hour of audio at an introductory price through year-end. It also launched MAI-Voice-2.1 and MAI-Voice-2.1-Flash text-to-speech models that speak 23 languages in one voice; Flash has 150ms latency and costs $15 per million characters, down from $22. The models are available via Microsoft Foundry, MAI Playground and OpenRouter.