BenchmarkApril 2, 20261 min read
Microsoft Unveils Breakthrough Speech-to-Text Model, Outperforming Rivals by a Significant Margin
Microsoft's new MAI-Transcribe-1 model achieves the lowest word error rate on the FLEURS benchmark, surpassing competitors like Scribe v2 and Whisper-large-V3, and offers a significant speed boost at an affordable price. This development is set to revolutionize speech-to-text capabilities for developers and businesses alike, with far-reaching implications for voice agents, transcription services, and more.
MAI-Transcribe-1 converts speech to text quickly and accurately in 25 languages, even with background noise. Microsoft is already using the model in its own products. The article Microsoft's MAI-Transcribe-1 runs 2.5x faster than its predecessor at $0.36 per audio hour appeared first on The Decoder.