Microsoft AI Releases MAI-Transcribe-2 Transcription Model

Microsoft AI released MAI-Transcribe-2, its latest speech transcription model. Artificial Analysis reports the model reached 2.0% AA-WER for a #2 ranking across datasets including AA-AgentTalk, VoxPopuli-Cleaned-AA, and Earnings22-Cleaned-AA. It also posted the #2 speed at a 411x factor and sits on the accuracy-speed Pareto frontier. Pricing through Microsoft Foundry is listed at $1.67 per 1,000 minutes of audio, lower than several other high-accuracy options in the comparison. Full results appear on the Artificial Analysis non-streaming Speech to Text leaderboard.