Microsoft AI's 2 New Speech Models, Do They Sound Human?

Talking to a voice agent can still feel like talking to a bot. Two things fix that. The agent has to understand you while you're talking, and it has to sound like a person when it answers. Microsoft AI has two new speech models for those two jobs: MAI-Transcribe-2-Streaming for listening and MAI-Voice-2.1-Flash for speaking. Sign up for LiveKit Cloud: https://cloud.livekit.io/signup?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=auffUDCMkoE In this video, I build a customer support voice agent on both models with LiveKit Agents. Then I call it about a laptop that didn't show up. The call is real and uncut. You'll also hear Harper, the voice in the demo, speak Indonesian and Portuguese. At the end I show the pipeline and the code, which is one AgentSession. The demo uses Gemma 4 through LiveKit Inference as the LLM, with the Microsoft AI plugin handling speech-to-text and text-to-speech. 📚 Resources 📚 Microsoft AI STT plugin docs: https://docs.livekit.io/agents/models/stt/microsoft-ai/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=auffUDCMkoE Microsoft AI TTS plugin docs: https://docs.livekit.io/agents/models/tts/microsoft-ai/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=auffUDCMkoE Agent docs: https://docs.livekit.io/agents/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=auffUDCMkoE 🤝 Join the Community: https://community.livekit.io/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=auffUDCMkoE #livekit #ai #voiceai