Hey everyone.
Imagine you're in a massive meeting with
20 people.
Everyone is talking at once and you need
a perfect transcript that knows exactly
who said what the moment they say it.
Usually, that's expensive and slow.
But today, Meta just changed the game.
Meta Superintelligence Labs has
officially launched News Voice
Transcribe.
This is a brand new audio perception
model that handles streaming
transcription and speaker diarization
for over 20 people simultaneously.
The most shocking part?
The price.
Meta is offering this for just 18 cents
per hour of processed audio.
To put that in perspective, competitors
like AssemblyAI or Deepgram can cost two
to three times more when you add in
speaker identification.
Muse isn't just cheap, it's smart.
It processes audio in tiny
80-millisecond chunks using a trick
called adaptive delay.
It can wait a split second if the speech
is messy or commit immediately if it's
clear.
It supports over 70 languages and
handles things like code-switching where
people jump between languages in one
sentence.
Now, is it a world record?
Not quite.
Speechmatics can handle up to 100
speakers and Amazon Transcribe handles
30.
But for most businesses building meeting
assistants or call analytics, 20
speakers is plenty.
Plus, Meta is topping the charts for
accuracy.
In recent benchmarks from Artificial
Analysis, Muse hit a 3.1% word error
rate beating out rivals like OpenAI,
Google, and ElevenLabs.
It's also leading in diarization
accuracy, which means fewer mistakes in
figuring out who is talking.
While there are some tradeoffs like no
word-level timestamps yet, the value
here is incredible.
For enterprise developers, this is a
massive win for building live AI
assistants and meeting tools without
breaking the bank.
Meta is making high-end voice AI
accessible to everyone.
What do you think?
Is 18 cents an hour low enough to change
how your company uses AI?
Let me know in the comments.
Like, subscribe, and turn on
notifications.