A blurred, abstract landscape featuring soft green and yellow hues, with hints of mountains and a field under a hazy sky. The image has a dreamy, painterly effect.

MAI-Transcribe-2

Rapidly turn any audio into highly accurate text transcripts. A single model for multilingual transcription in any environment, with diarisation, timestamps, domain biasing all built-in.

Features

MAI-Transcribe-2 delivers consistent, high-accuracy transcription across languages, accents, and challenging audio conditions.

And, as of today, MAI‑Transcribe‑2‑Streaming offers real-time transcription, with no hefty latency tradeoff.

Together, they give users fast, fluid, and conversational building blocks with no compromise on accuracy or voice quality.

Understands your industry terminology

Uses keyword biasing to tailor transcription to domain-specific terminology, making it ready for captions, call analysis, accessibility, and content workflows out of the box.
Agent 00:04

Thank you for calling Pharmacy Support. My name is Jordan. How can I help you today?

HCP 00:10

Hi. I was charged twice for my Lisinopril refill last week and I haven’t heard anything about a refund.

Agent 00:19

I’m sorry about that. It looks like the auto-refill processed at the same time as your manual order. I’ll reverse the extra charge now — you should see it back within 3 to 5 business days.

HCP 00:31

That makes sense. Thank you for sorting that out.

Daniel 00:06

Let’s start with the Q3 budget review. I sent the updated spreadsheet this morning — did everyone get a chance to look at it?

Sara 00:14

I did. The travel line item looks high to me. Can we revisit that before we approve?

Daniel 00:21

Sure. I think two of those trips could be moved to virtual. That would bring it in line with last quarter.

Sara 00:29

Agreed. Let’s flag it for revision before the final sign-off.

Anton Johnson
00:05

This is Dr. Johnson. Dictating notes for a follow-up visit with Marcus T., 47-year-old male.

00:12

Patient presents with persistent lower back pain, three weeks in duration. Reports morning stiffness that improves with movement. No radiation to the legs.

00:24

Assessment: likely mechanical lower back pain. No red flag symptoms. Contributing factors include posture and sedentary work habits.

00:32

Plan: referral to physical therapy, six sessions. Ibuprofen 400 milligrams as needed with food. Follow up in four weeks or sooner if symptoms worsen. End note.

Leading accuracy across 60 languages

State-of-the-art transcription quality across a globally diverse set of languages and speaking styles, with very precise automatic language detection.

Deutsch

Chinese 中文

Italiano

Hindi हिन्दी

English (Australian)

Français

Español

English

Japanese 日本語

Português

Performance

Industry leading accuracy

MAI-Transcribe-2.1 achieved the lowest Word Error Rate among leading speech-to-text models. On FLEURS across 60 languages, it outperformed Whisper-Large-V3, GPT-Transcribe, ScribeV2 and Gemini 3.5 transcribe

Version Comparison

MAI-Transcribe-2

  • Position in Artificial Analysis Accuracy Leaderboard #2
  • Time to first partial N/A
  • Latency end-to-end 1hr audio → 10 sec
  • Languages 60
  • Real-time streaming NO
  • Transcription Style YES – Clean and Verbatim
  • Word-level timestamps YES
  • Contextual / keyword biasing YES
  • Diarization YES
  • Pricing (introductory) $0.10 per hour of audio
View docs

MAI-Transcribe-2-Streaming

  • Position in Artificial Analysis Accuracy Leaderboard #1
  • Time to first partial ~120ms
  • Latency end-to-end ~128ms
  • Languages 60
  • Real-time streaming YES
  • Transcription Style NO
  • Word-level timestamps NO
  • Contextual / keyword biasing NO
  • Diarization NO
  • Pricing (introductory) $0.54 per hour of audio
View docs

MAI-Transcribe-1.5

  • Position in Artificial Analysis Accuracy Leaderboard #3
  • Time to first partial N/A
  • Latency (end-to-end) 1hr audio → 20 sec
  • Languages 43
  • Real-time streaming NO
  • Transcription Style NO
  • Word-level timestamps NO
  • Contextual / keyword biasing NO
  • Diarization NO
  • Pricing $0.36 per hour of audio
View docs

Try MAI-Transcribe-2

MAI Playground

Experiment with all other MAI models.
Try in playground

Microsoft Foundry (Azure Speech)

Turn every spoken word into accurate, searchable text instantly.
Try it in Foundry
English (United States)
Your Privacy Choices Opt-Out Icon Your Privacy Choices
Consumer Health Privacy Sitemap Contact Microsoft Privacy Manage cookies Terms of use Trademarks Safety & eco Recycling About our ads