Microsoft's MAI-Transcribe-2 Undercuts OpenAI, Google and ElevenLabs (venturebeat.com)

Microsoft AI has launched MAI-Transcribe-2, a new speech-recognition model that significantly undercuts competitors like OpenAI, Google, and ElevenLabs by offering faster and more accurate transcription at an attractive price of just 10 cents per hour of audio, compared to its predecessor's $0.36. This 72% price reduction could lead to substantial cost savings for enterprises, such as banks and telecoms, which can see their annual transcription bills drop from $36,000 to $10,000, effectively making transcription a non-negotiable expense. The model supports transcription in 60 languages and boasts advanced features like speaker diarization, word-level timestamps, and automatic language identification—all included in the base product. The release is significant as it marks Microsoft's rapid advancement in developing its AI models independently, moving away from reliance on OpenAI after a substantial investment. MAI-Transcribe-2 illustrates Microsoft’s strategy to enhance its AI capabilities, particularly useful for business applications, while maintaining competitive superiority. The model is also designed for real-world conditions, handling messy audio such as background noise and overlapping speech, thereby catering directly to the needs of enterprise clients. Microsoft's ability to release three iterations in just five months showcases its commitment to innovation and cost-effectiveness in the AI space, potentially reshaping the transcription market landscape.