Microsoft AI’s MAI-Transcribe-2 undercuts OpenAI, Google and ElevenLabs on price and speed
Microsoft AI on Thursday released MAI-Transcribe-2, a speech-recognition model the company claims is faster, more accurate, and cheaper than those offered by OpenAI, Google, or ElevenLabs. The company priced it at just 10 cents per hour of audio.
This significant drop is worth noting, as Microsoft AI initially charged $0.36 per hour just five months ago. This new price cuts that cost by roughly 72%, saving a substantial amount for enterprises processing large volumes of call-center audio.
The release marks a strategic shift for Microsoft, as it builds its own AI models and integrates them into its products, competing directly with OpenAI without relying on its $13 billion investment.
What MAI-Transcribe-2 offers
- Support for 60 languages, an increase from 43 in June's MAI-Transcribe-1.5 and 25 in the original release.
- Speaker diarization, sorting out speaker contributions in multi-person recordings.
- Word-level timestamps, enabling search, editing, and alignment with video.
- Keyword biasing, filtering out domain jargon.
- Automatic language identification.
- Configurable output styles, including "verbatim" and "clean" modes.
- Code switching to handle conversations that switch languages mid-sentence.
All these features, which were once offered as add-ons or premium services, are now included for just 10 cents per hour.
Performance Benchmarks
Microsoft makes three performance claims:
- MAI-Transcribe-2 ranks first on FLEURS, a benchmark with a 5.2% average word error rate across 60 languages.
- It surpasses competitors in terms of speed and accuracy.
- The model is trained on messy audio, demonstrating its capability in real-world scenarios.