Microsoft launched three new voice AI models on October 1 under its MAI brand, expanding its push into speech technology.

The new models are MAI-Transcribe-2-Streaming, MAI-Voice-2.1 and MAI-Voice-2.1-Flash. They cover transcription and text-to-speech services for developers.

The transcription model supports 60 languages and delivers partial results in about 100 milliseconds, making it suitable for live captioning and real-time conversations. Microsoft prices it at $0.54 per audio hour.

The text-to-speech models support 23 languages and are priced at $22 per million characters. They are designed for voice assistants, audiobooks and accessibility tools.

Microsoft is positioning the MAI voice line against rivals including OpenAI's Whisper and Google's speech APIs. The company says the new models balance accuracy, speed and cost.

Voice AI has become a crowded market as businesses add speech interfaces to customer service, healthcare and education products. Pricing has emerged as a key battleground.

Developers can access the models through Microsoft's cloud platform, with the faster Flash variant aimed at applications where latency matters most.