Microsoft MAI-Transcribe-1.5 Tops Speech Recognition Benchmarks

.@ArtificialAnalysis’ graph shows that MAI-Transcribe-1.5 is in a league of its own

@mustafasuleyman Hi Mustafa, did you know that Richard Socher acquired http://Recursive.si to build SuperIntelligence? We are rapidly shifting into a new era of SuperIntelligence (SI).

Even Geoffrey Hinton recently stated: 'SuperIntelligence is coming'

We know Microsoft owned Copilot•si

@mustafasuleyman Good addition to portfolio

@mustafasuleyman Given the investments Microsoft Research put into voice for decades I’m not surprised.

@mustafasuleyman Whisper just lost its default status in every voice pipeline.

@venky4a @mustafasuleyman Any thoughts on Cerebras Systems?

@mustafasuleyman MAI-Transcribe-1 dominating is cool, now who's figuring out how to run it on consumer hardware without melting it? ⚡️

@mustafasuleyman must be nice to be the only dot in the attractive quadrant

speed + accuracy is a rare combo in this space

@mustafasuleyman I love how the green is labeled as ‘Most attractive quadrant’.

@mustafasuleyman the separation is clear. curious how it handles code-switching and heavy-accent edge cases. whisper still struggles with those in production

@mustafasuleyman wait, is it actually usable or just benchmark king

@mustafasuleyman Another model with high benchmarks, wake me when it actually does something useful.

@mustafasuleyman What's the main differences with the v1?

@mustafasuleyman Mustafa, this is impressive. At this point there is no comparison

@mustafasuleyman This kind of benchmark comparison is super helpful. Seeing how models stack up head-to-head like this really clarifies where the state of the art is.

@mustafasuleyman Don't see it in my enterprise foundry yet.

@mustafasuleyman straight up, and their models arent even close to that same quadrant yet

@mustafasuleyman Does it work on accented speech and real-world noise?

That's where my voice bots struggle most. Is this production-ready for Indian accents + street sounds, or just clean audio?

@mustafasuleyman being in its own tier on a log scale chart is the strongest flex in benchmarks

@mustafasuleyman When the chart needs a whole extra zip code just to fit your data point, massive achievement! 🌐⚡