A note-taking app for class

One voice in a noisy room.
Every word of it.

Lecture halls are a cocktail party: the professor across the room, people talking, chairs, doors. Add a heavy accent and most apps give up. Verbat records the room, follows the voice you choose, and writes every word. After class, it turns the lecture into notes you'll actually reread.

Free during the pilot. iOS first. One email when it's ready.

Live captions
Professor

Students

The problems

Two problems. One app.

01

The noisy room

The cocktail-party problem. The professor is five meters from your phone, someone drops a backpack, the room echoes, people whisper. Your phone hears all of it, mixed into one wall of sound. Verbat separates the voice you care about from the room.

02

The heavy accent

A professor's English carries an accent you're still learning to hear. Most transcription apps guess, and the guess lands wrong. By the time you decode the sentence, the lecture has moved on. Verbat is built for accented English, and we publish the numbers.

The solution

A notepad that listens.

Verbat sits in your pocket during class. It hears everything, writes it down in real time, and keeps each voice separate so you can always tell who said what.

01

It records the lecture

The whole class, start to finish. You never have to write fast again.

02

Every word appears as live text

Read along while the professor speaks, through the noise and the accent.

03

Every voice, sorted after class

Each speaker gets their own clean transcript. Read just the professor's words, without the room.

04

It writes your notes for you

After class, the lecture becomes clean, organized notes: the key ideas, in order, ready to review. Not a transcript dump.

Hear it yourself

Same room. Same noise.
Different result.

A real lecture, recorded the way it actually sounds: professor at a distance, room noise, accented English. During class it's a wall of sound, and most apps leave it that way forever. After class, Verbat separates the voices and writes each speaker clean. Here is the same moment, as typical apps left it and as Verbat returns it.

0:000:20

The recording: a real Korean-accented lecture, captured from the back of the room. The professor is far from the mic. The room is alive.

Typical apps

Verbat

Same audio, same moment, same conditions. Typical apps is the best result from the other engines we tested. Full numbers below.

Every voice, separated.

This recording has two voices talking over each other. Live captions mix them into one mess, and most apps leave it that way. After class, Verbat separates the recording by speaker: tap a track to hear one voice alone.

0:000:22

Tap a voice to hear it alone. Voices are separated cleanly, one track per speaker.

Not a live effect: separation runs after the recording finishes.

After class

The lecture becomes notes.

Nobody rereads a transcript dump. After class, Verbat separates every voice, writes each speaker clean, and condenses the lecture into the notes you actually want: the key ideas, organized, ready before your next exam.

The raw transcript

Your notes

The lecture is condensing.

Illustrative example. Every recording ends with this.

The benchmark

Measured. Published.

No marketing numbers. Every score below comes from the same 20-second clips, run through real classroom acoustics and real classroom noise at three levels. Lower is better: 0% means every word right, 100% means word soup.

AccentVery loud roomLoud roomModerate noise
Korean EnglishReal lecture recording
86.7%
86.7%
23.3%
Vietnamese EnglishSynthesized accented voice
80.8%
69.2%
36.5%
Native EnglishReal lecture recording
32.4%
15.5%
9.9%
Indian EnglishReal lecture recording
15.9%
20.5%
18.2%
Chinese EnglishSynthesized accented voice
3.8%
1.9%
0.0%
Spanish EnglishSynthesized accented voice
1.9%
0.0%
0.0%

Honest note: these scores come from our far-field test rig: real room acoustics, real classroom babble, phone-style capture. The test set mixes real lecture recordings with synthesized accented voices where real recordings are not available yet, marked above. Real classrooms are the next test.

Methodology: same 20-second clips, three noise levels, measured August 2026. Typical apps in the demo above = the best of the other engines we tested under identical conditions. No engine names here: you don't buy engines, you buy your lecture back.

These are the first published numbers, not a ceiling: the pipeline improves continuously, we re-run the benchmark every time it does, and the scores on this page get updated as they get better.

The waitlist

Be first in line.

Verbat ships in early access. Sign up and we'll reach out with a short survey, then beta invites.

No spam. Unsubscribe anytime.

You're on the list.

We'll reach out soon.