Skip to content
Doc2X

Turn a recording into notes with times.

Upload a lecture, a talk or a meeting as video or audio. It comes back as a transcript, sentence by sentence, and notes in sections: a heading, a few points and the time each part starts at. For a video, each section shows what was on screen.

Make notes from a recording

What was on screen 0:34 into the sample lecture: a slide headed "The learning rate is the step size".

0:34The learning rate

  • The learning rate is the size of each step.
  • If the learning rate is too small, training takes forever, because every step barely moves the parameters.
Section 2 of 4From a 1:49 lecture
Recording
A video or an audio file from your device
Spoken
English or Chinese: detected, or you say which
Notes in
The language spoken, or one of 12 you choose
Open Notes

Free plan: 50 credits every month, no card needed. That is 50 minutes of recordings.

Nothing at hand? See the notes a two-minute lecture came back as

From
A video or an audio file (MP4, MOV, WebM, MP3, M4A, WAV and the like) of up to 180 minutes and 2 GB
Back
A transcript with times, notes in sections, a summary, and for a video a picture of the screen for each section
Cost
1 credit for each minute of the recording, charged when the notes are done

One lecture, its notes

Two minutes of lecture, as they came back.

These are the notes for a short lecture on gradient descent: five slides and a voice. The sections, their times, the points and the pictures are what came back for the file. The transcript is under them.

  • 01Each section starts within a second of the slide that opens it: 0:00, 0:34, 1:08, 1:31.
  • 02Numbers spoken as words are numbers in the notes: “zero point zero one” is 0.01.
  • 03Of the 306 words spoken, 305 came back as spoken. “Plain” came back as “plane”, and it is left in.

Notes1:49 4 sections 15 sentences

Gradient Descent, Learning Rate, and Momentum

The lecture introduces gradient descent as a way to minimize a loss function that measures how wrong a model is. It explains that the learning rate sets the step size: too small and training takes forever, while too large is also problematic. Momentum addresses the zigzagging of plain gradient descent in narrow valleys by keeping a running average of past gradients, called the velocity. The main conclusions are that gradient descent follows the slope downhill, and the learning rate must be neither too small nor too large.

  1. What was on screen 0:00 into the lecture, when the section “Introduction and the idea of gradient descent” starts.

    0:00Introduction and the idea of gradient descent

    • The lecture covers what gradient descent is, how the learning rate changes its behavior, and why momentum helps.
    • We have a function that measures how wrong our model is, and we call it the loss.
    • Gradient descent looks at the slope of the loss at the current parameters and takes a small step downhill.
    • We repeat this until the loss stops getting smaller.
    • Each step uses the gradient, which is the vector of partial derivatives of the loss with respect to every parameter.
  2. What was on screen 0:34 into the lecture, when the section “The learning rate” starts.

    0:34The learning rate

    • The learning rate is the size of each step.
    • If the learning rate is too small, training takes forever, because every step barely moves the parameters.
    • If the learning rate is too large, the steps overshoot the bottom of the valley and the loss can even grow.
    • A common starting value is 0.01, and people often reduce it by a factor of ten when the loss stops improving.
    • In the experiment, a learning rate of 0.1 diverged after twenty steps, while 0.01 reached a loss of 0.3 after two hundred steps.
  3. What was on screen 1:08 into the lecture, when the section “Momentum” starts.

    1:08Momentum

    • Plain gradient descent zigzags in narrow valleys.
    • Momentum fixes this by keeping a running average of past gradients, which we call the velocity.
    • The update adds ninety percent of the previous velocity to the new gradient in the narrow valley.
    • With momentum, the same loss was reached in sixty steps instead of two hundred.
  4. What was on screen 1:31 into the lecture, when the section “Summary and outlook” starts.

    1:31Summary and outlook

    • Gradient descent follows the slope downhill.
    • The learning rate controls the step size and must be neither too small nor too large.
    • Momentum smooths the path and speeds up training.
    • Next week the lecture will look at adaptive methods such as Adam.
The transcript: 15 sentences, each with its time
  1. 0:00Welcome to this short lecture on gradient descent. Today we will cover three things. What the method is?
  2. 0:06How the learning rate changes its behavior and why momentum helps.
  3. 0:11Let us start with the idea. We have a function that measures how wrong our model is.
  4. 0:16We call it the loss.
  5. 0:17Gradient descent looks at the slope of the loss at the current parameters and takes a small step downhill. We repeat this until the loss stops getting smaller. Each step uses the gradient, which is the vector of partial derivatives of the loss with respect to every parameter.
  6. 0:34Now the learning rate.
  7. 0:35The learning rate is the size of each step. If it is too small, training takes forever, because every step barely moves the parameters. If it is too large, the steps overshoot the bottom of the valley and the loss can even grow.
  8. 0:49A common starting value is zero point zero one, and people often reduce it by a factor of ten when the loss stops improving. In our experiment, a learning rate of zero point one diverged after twenty steps.
  9. 1:03While zero point zero one reached a loss of zero point three after two hundred steps.
  10. 1:08The third topic is momentum plane gradient descent zigzags in narrow valleys.
  11. 1:14Momentum fixes this by keeping a running average of past gradients.
  12. 1:18Which we call the velocity. The update adds ninety percent of the previous velocity to the new gradient in the narrow valley. Example, momentum reached the same loss in sixty steps instead of two hundred.
  13. 1:31To summarize.
  14. 1:32Gradient descent follows the slope downhill. The learning rate controls the step size and must be neither too small nor too large, and momentum smooths the path and speeds up training. Next week we will look at adaptive methods such as adam.
  15. 1:47Thank you for listening.

Not a mock-up. These are the notes Doc2X returned on 20 November 2026 for a recording of our own: a lecture of 1 minute 49 seconds, five slides read out by a computer voice. Headings, points, times and transcript are as they came, slips included, and the pictures are the ones it took of the screen. It cost 2 credits and was done 14 seconds after the upload. Open the recording that went in.

Why you will find it again

Notes that know where they came from.

  1. Sections1:49 of lecture
    1. 0:00Introduction and the idea of gradient descentThe lecture covers what gradient descent is, how the learning rate changes its behavior, and why momentum helps.
    2. 0:34The learning rateThe learning rate is the size of each step.
    3. 1:08MomentumPlain gradient descent zigzags in narrow valleys.
    4. 1:31Summary and outlookGradient descent follows the slope downhill.
    The sections that came back, each with the time it starts at and its first point

    01

    Sections with the time they start at

    The recording is divided where the subject changes. Every section has a heading, a few points and its starting time, and in the app a click on a section moves the player there.

  2. Transcript15 sentences
    1. 0:00Welcome to this short lecture on gradient descent. Today we will cover three things. What the method is?
    2. 0:06How the learning rate changes its behavior and why momentum helps.
    3. 0:11Let us start with the idea. We have a function that measures how wrong our model is.
    4. 0:16We call it the loss.
    5. 0:17Gradient descent looks at the slope of the loss at the current parameters and takes a small step downhill. We repeat this until the loss stops getting smaller. Each step uses the gradient, which is the vector of partial derivatives of the loss with respect to every parameter.
    6. 0:34Now the learning rate.
    The first six sentences of the transcript, as they came

    02

    Every sentence, with its time

    Under the notes is the transcript, sentence by sentence. It can be searched, a click on a sentence plays the recording from there, and notes and transcript download together as one Markdown file.

  3. The screen 0:00 into the lecture, at the start of the section “Introduction and the idea of gradient descent”.0:00
    The screen 0:34 into the lecture, at the start of the section “The learning rate”.0:34
    The screen 1:08 into the lecture, at the start of the section “Momentum”.1:08
    The screen 1:31 into the lecture, at the start of the section “Summary and outlook”.1:31
    The pictures that came back: the screen at the start of each of the 4 sections

    03

    What was on screen

    For a video, each section comes with a picture of the screen at the moment it starts: the slide, the board, the shared window. An audio file gets the notes without the pictures.

  4. Notes inAs spoken

    0:34The learning rate

    • The learning rate is the size of each step.
    • If the learning rate is too small, training takes forever, because every step barely moves the parameters.
    • If the learning rate is too large, the steps overshoot the bottom of the valley and the loss can even grow.
    Notes inSpanish

    0:34Tasa de aprendizaje

    • La tasa de aprendizaje es el tamaño de cada paso.
    • Si es demasiado pequeña, el entrenamiento tarda muchísimo porque cada paso apenas mueve los parámetros.
    • Si es demasiado grande, los pasos se pasan del fondo del valle y la pérdida puede incluso aumentar.

    The same recording, asked for twice: one section of each set of notes

    04

    Notes in your language

    The speech has to be English or Chinese. The notes can be written in the language spoken or in one of twelve you choose. The transcript stays as it was spoken.

How it works

Three steps from recording to notes.

  1. 01

    Choose the recording

    Open Notes and choose a video or an audio file from your device: up to 180 minutes and 2 GB. The page says what it will cost, by the minute, before you start.

  2. 02

    Say what the notes are in

    English or Chinese speech is detected, or you set it. Choose the language for the notes, click Make notes, and keep the page open while the file uploads.

  3. 03

    Read, jump, copy

    The notes open by themselves when they are done. Jump to a section, search the transcript, copy the notes, or download them with the transcript as a Markdown file.

The fine print

What it takes and what it gives back.

Better to know before you upload. The last list is what it does not do today.

Files
A video or an audio file: MP4, MOV, WebM, MP3, M4A, WAV and the like
Length and size
Up to 180 minutes and 2 GB a recording
Speech
English or Chinese. It is detected, or you say which
Notes in
The language spoken, or one of 12 you choose: Arabic, Chinese (Simplified or Traditional), English, French, German, Italian, Japanese, Korean, Portuguese (Brazil), Russian or Spanish
What comes back
A title and a summary of the whole; sections, each with a heading, a few points and the time it starts at; the transcript, sentence by sentence, each with its time
Pictures
For a video, a picture of the screen at the start of each section
On the page
The recording plays beside the notes, from the section or the sentence you click. The transcript can be searched; the notes can be copied, or downloaded with the transcript as a Markdown file
How long it takes
The lecture on this page, 1 minute 49 seconds long, was done 14 seconds after it had arrived. A large file takes its time to upload: keep the page open until it is through
Cost
1 credit for each minute of the recording that has begun, on every plan, charged when the notes are done. Notes that could not be made cost nothing
Storage
Recordings and notes are deleted after 31 days. You can delete them sooner

Not a fit yet

  • A link to YouTube or to another site. Upload the file itself.
  • Speech in languages other than English and Chinese.
  • Who is speaking. The transcript is one stream of sentences and does not name speakers.
  • A transcript without a slip. A word is misheard now and then, and a sentence can end a few words early or late; the notes are written from the transcript, so a slip there can show in them.
  • Changing the notes on the page. Copy or download them to work on them.

Questions

Questions people ask first.

Something else on your mind? The help center goes deeper.

How do I turn a video into notes?

Open Notes, choose the video or audio file, pick the language for the notes and click Make notes. Keep the page open while the file uploads. The notes open when they are done: a summary, sections with the time each starts at, and the transcript.

Is it free?

The free plan gives you 50 credits every month and needs no card. Notes cost 1 credit for each minute of the recording, so a month of the free plan covers 50 minutes. The page says what a recording will cost before you start.

Does it work with a YouTube link?

Not yet. Upload the file: a video or an audio file of up to 180 minutes and 2 GB.

Which languages does it understand?

English and Chinese speech. The notes can be written in the language spoken or in one of 12 you choose, among them Spanish, French, German, Japanese, Korean and Arabic. The transcript stays in the language spoken.

Does it work with audio only?

Yes. An audio file gets the transcript, the sections and the summary. The pictures of the screen come with videos only.

How accurate is the transcript?

Good enough to find your way in a recording and to quote from once you have checked. In the lecture on this page, 306 words read by a computer voice, one word came back wrong (“plane” for “plain”). The other 305 are all there, though the sentences are not always cut where the speaker paused. Names, numbers and technical terms are where to look twice.

Can I use it for meetings?

A recorded meeting works like any other recording. The transcript does not say who is speaking, so the notes tell you what was said and when, not who said it.

What happens to my recording?

It is uploaded over an encrypted connection, and recordings and notes are deleted automatically after 31 days; you can delete them sooner. Unless you explicitly agree, your files are not used to train AI models offered to the public or to other customers.

Bring the recording you would have to watch twice.

Start on the free plan. No card needed.