
Microsoft MAI
Microsoft AI speech models: MAI Transcribe 2, MAI Voice 2.1 and MAI Realtime 1 speech-to-speech.

About Microsoft MAI
Microsoft MAI: Speech Models from Microsoft AI
Microsoft MAI is the family of speech models built by Microsoft AI: MAI Transcribe 2 for speech-to-text, MAI Voice 2.1 for text-to-speech and MAI Realtime 1 for live speech-to-speech conversations. Developers use them through Microsoft Foundry, with demos in MAI Playground and MAI Labs.
Key Features
MAI Transcribe 2: Speech-to-text in 60 languages, with automatic language identification, code-switching such as Hinglish and Spanglish, diarization, word-level timestamps and keyword biasing for domain terms.
Batch and Streaming: Batch transcription processes an hour of audio in about 10 seconds of model inference. MAI Transcribe 2 Streaming transcribes live speech with about 128 ms of model inference latency, without diarization, timestamps or biasing.
Noisy Audio: Built for background noise, variable audio quality and mixed-language speech in calls, meetings and voicemail.
MAI Voice 2.1 and 2.1 Flash: Expressive text-to-speech in 23 languages with granular emotion control. Voice 2.1 is tuned for fidelity (about 550 ms model inference), Flash for speed (about 45 ms).
Zero-shot Voice Prompting: A custom voice from 5 to 60 seconds of reference audio, with no fine-tuning, and consent guardrails so only authorized voices can be used.
MAI Realtime 1: A speech-to-speech model for live conversations, with turn-taking that handles interruptions, recognition through background noise, tool use and personas set by prompting.
Reasoning on Demand: A fast system keeps the conversation going and calls a stronger reasoning engine when a request needs it.
OpenAI Realtime API Compatibility: MAI Realtime 1 works with the OpenAI Realtime API, so existing voice apps can be tested and moved over.
Use Cases
Contact Centers: Call transcription and analysis, and voice agents that act on enterprise systems.
Healthcare Documentation: Clinical transcription: Dragon Copilot Transcribe is built on MAI Transcribe 2.
Captions and Meetings: Multilingual captions, meeting transcripts and accessibility workflows.
Branded Voices: Narration and product voices in 23 languages from a short reference clip.
Pricing
MAI Transcribe 2: $0.10 per hour of audio for batch and $0.54 per hour for streaming, both limited-time prices.
MAI Voice 2.1: $22 per 1M characters. MAI Voice 2.1 Flash: $15 per 1M characters.
MAI Realtime 1: $0.01 per audio minute, with reasoning billed at $1 / $5 per 1M input / output tokens. Limited-access preview, not billed until public preview.
Getting Started
Website: https://microsoft.ai/models/
MAI Realtime 1: https://microsoft.ai/models/mai-realtime-1/
MAI Transcribe 2: https://microsoft.ai/models/mai-transcribe-2/
MAI Voice 2: https://microsoft.ai/models/mai-voice-2/
Microsoft MAI covers transcription, speech synthesis and live speech-to-speech from one vendor, sold through Microsoft Foundry.
