To extract the sound, open the recording in the converter. This guide covers where each recorder saves the file, what to check before converting, and what the converter does with a recording that holds more than one audio track.
A screen recording is not an MP3
The capture is a video container — MP4, MOV, MKV or WebM — with a picture track and at least one audio track. Extracting audio writes a new sound file. It does not convert the recording itself into an MP3, and it does not fetch a file from a website. Open a file already on this device in the converter, then press Convert and Save file.
Find the recording on your device first
The recorder decides the container, and the container decides which export paths are open. Look at the file before opening it here.
- Mac. Apple documents that screen recordings made with the Screenshot tool “are saved as .mov files”, on the desktop by default, with names that begin with “Screen Recording” and include the date and time. MOV with AAC audio is the case where an unchanged M4A copy is usually possible. (Apple: take a screenshot or screen recording on Mac)
- Windows, Xbox Game Bar. Microsoft documents that Windows key + Alt + R starts and stops a recording, Windows key + Alt + M toggles the microphone, and the result is an MP4 file under Videos in a folder called Captures. (Microsoft: record your screen with Xbox Game Bar)
- iPhone. Apple says a screen recording is found in the Photos app, in your Library, and that the microphone is turned on by touching and holding the Record button before starting. If you pick it from Photos in the browser, iOS may prepare a converted copy first; saving the recording to Files and choosing it through Browse avoids that step. (Apple: record the screen on iPhone)
- Android. Google’s help page puts screen recordings in Photos under Collections, On this device, Movies, and the recorder asks whether to record audio before it starts. Exact menus differ between phone makers. (Google: take a screenshot or record your screen on Android)
After the file opens, the File check panel under the waveform shows what the recorder actually wrote: the container, every audio track with its name, codec, channels, sample rate and measured bitrate, which export path the current choice will take, and the expected size of each output. If the recording has no audio track, the page says so before any conversion starts, which usually means audio capture was off when recording. The test results further down list what that panel reported for six test files and how long each export took.
What a two-track recording looks like: the sample
The converter has an Open the sample recording button under the drop zone, for when there is no recording at hand. The sample is a generated test file, not the output of a real recorder: 20 seconds of Matroska, 474,107 bytes, made with FFmpeg from a 320×180 test pattern and two audio tracks. Track 1, named System audio, is AAC stereo encoded at 96 kbps and carries a 220 Hz tone with quiet pink noise. Track 2, named Microphone, is Opus mono encoded with a 32 kbps target and carries a 440 Hz tone that sounds for 1.8 seconds out of every 3, standing in for speech with pauses. Its SHA-256 is 8880b2c704b8aa42410cbc22e135b36ebe845c1ef3fce03172b2aa54503f0c41.
Opened in desktop Chrome on 13 October 2026, File check listed both tracks by name with language, codec, channels and the 48 kHz sample rate, and measured 96 kbps for System audio and 46 kbps for Microphone. The second figure is above the 32 kbps the file was encoded with, because File check adds up the packets that are actually in the file rather than reading a setting. These are the files the converter saved:
| Track | Output | Saved size |
|---|---|---|
| 1. System audio (AAC stereo) | MP3, 192 kbps | 481,704 bytes |
| 1. System audio (AAC stereo) | M4A, unchanged copy | 250,276 bytes |
| 1. System audio (AAC stereo) | WAV, 16-bit | 3,844,076 bytes |
| 2. Microphone (Opus mono) | MP3, 192 kbps | 481,128 bytes |
| 2. Microphone (Opus mono) | M4A, new AAC encode | 485,294 bytes |
| 2. Microphone (Opus mono) | WAV, 16-bit | 1,921,340 bytes |
Three things in that table apply to real recordings too. The MP3 is the same size for both tracks, because the converter has one MP3 rate and a mono source does not lower it. The unchanged M4A copy is offered for the AAC track and not for the Opus track, which has to be encoded again, so its M4A is about twice the size of the copy. WAV halves for the mono track, because WAV size is set by channels, sample rate and duration, not by how the source was compressed.
The input file is not copied into memory
A long screen capture is handled differently from a conventional whole-file converter. PrivateMP3 asks the local file for ranges as they are needed. It does not decode the picture frames. For the waveform, decoded audio is reduced to minimum and maximum peaks for time buckets rather than retained as full PCM. This makes the page useful for a lecture you recorded, a meeting captured with permission, a tutorial you made or another local screen recording where only the sound is needed.
The output can be the memory problem
The completed audio still has to be stored somewhere. PrivateMP3 normally streams it into the Origin Private File System, browser-managed storage on the same device, and exposes it through Save file when finished. If that interface is unavailable, the worker falls back to an in-memory output buffer. A long result is more demanding on that fallback even though the source continues to be read in ranges.
Duration still costs time and storage
The browser must decode the audio timeline, build waveform peaks and encode every kept sample. Device speed, free storage, audio codec and output format therefore matter. The picture resolution is not sent to the audio encoder, but the container still has to be read; two recordings of equal duration should not be promised equal processing times.
Keep the tab open and the device awake. The page requests a screen wake lock while converting when the browser allows it, but system power controls can still suspend a background tab.
Check the tracks before converting
A capture may store microphone and system audio as separate tracks, or mix them into one. When several audio tracks exist, a picker appears with codec and channel information. PrivateMP3 can select one track. It cannot combine tracks, separate speakers from a mix or recover audio that the recorder never captured. An interrupted recording may also be missing packets or final container metadata.
Choose the output for the job
For a recording, the File check panel shows the size of MP3, M4A and WAV before Convert is pressed, and says whether M4A will be an unchanged copy. What each output keeps is set out in the size and output guide.
Remove an empty introduction, long break or unused ending before export. The output-duration estimate updates to reflect what remains. To check whether the result will fit a size limit before converting, use the size calculator. Removing time does not always make the file smaller; the guide to smaller lecture audio has the rule and the measurements.
Measured: six generated recordings through the converter
How long does it take to get the sound out of an hour of screen recording, and how big is the result? These are measured numbers from the shipped tool, not estimates.
What was tested
Six files were made with FFmpeg 9.0.1 from its built-in test sources: a moving 30 fps test pattern for the picture and a 220 Hz tone mixed with quiet pink noise for the sound. They are not captures from any particular recorder. Each one copies a container and track layout that screen recordings commonly use, so the results describe how this tool handles those layouts, not how a given recording app behaves.
| File | Picture | Audio as written | Size |
|---|---|---|---|
| A. 60 min MP4 | H.264, 1920x1080 | AAC, stereo, 48 kHz, 160 kbps | 4,388,816,051 bytes (4.39 GB) |
| B. 10 min MOV | H.264, 1920x1080 | AAC, stereo, 48 kHz, 128 kbps | 728,935,720 bytes |
| C. 10 min MKV | H.264, 1920x1080 | Track 1 named System audio: AAC stereo, default. Track 2 named Microphone: Opus mono | 736,844,134 bytes |
| D. 10 min WebM | VP8, 1280x720 | Opus, stereo, 48 kHz, 96 kbps requested | 81,204,433 bytes |
| E. 1 min MP4 | H.264, 1920x1080 | None | 71,877,497 bytes |
| F. 30 s MKV | H.264, 640x360 | AAC, 5.1 (six channels), 48 kHz | 7,999,438 bytes |
The built site was served locally and driven in Chrome 153.0.8010.53 on Windows 11, on an Intel Core i5-13600KF with 32 GB of memory, using Playwright. Advertising and analytics requests were blocked so the test did not count as visits. Every saved file was read back with ffprobe to confirm codec, sample rate, channels and duration. Run date: 5 October 2026.
What the File check panel reported
The panel appears under the editor once a file opens. Its bitrate is measured from the track's packets, because containers rarely store one.
| File | Panel summary | Ready to edit after |
|---|---|---|
| A | MP4, 1 audio track, 1 video track. AAC, stereo, 48 kHz, 160 kbps | 12.0 s |
| B | QuickTime File Format, 1 audio, 1 video. AAC, stereo, 48 kHz, 128 kbps | 1.3 s |
| C | Matroska, 2 audio, 1 video. 1. System audio, eng, AAC, stereo, 160 kbps, default. 2. Microphone, eng, Opus, mono, 73 kbps | 1.8 s |
| D | WebM, 1 audio, 1 video. Opus, stereo, 48 kHz, 69 kbps | 1.8 s |
| E | Refused at open: This file has no audio track, so there is nothing to extract. | 0.1 s |
| F | Matroska, 1 audio, 1 video. AAC, 6 ch, 48 kHz, 186 kbps | 0.2 s |
Ready means the waveform for the whole timeline was drawn. That time follows the length of the audio, not the size of the file: the 4.39 GB hour took 12 seconds, because the picture is never decoded. File D shows why the bitrate is measured. FFmpeg was asked for 96 kbps Opus, but Opus varies its rate with the signal, and the simple test sound needed only 69 kbps.
Export times and sizes
| File | Output | Time | Saved size | Panel estimate |
|---|---|---|---|---|
| A (60 min) | MP3 | 61.4 s | 86,400,744 bytes | About 86 MB |
| A | M4A, unchanged copy | 6.1 s | 72,676,105 bytes | About 72 MB |
| A | WAV | 29.4 s | 691,200,044 bytes | About 691 MB |
| B (10 min) | MP3 | 8.8 s | 14,400,744 bytes | About 14 MB |
| B | M4A, unchanged copy | 1.2 s | 9,739,957 bytes | About 9.6 MB |
| B | WAV | 5.0 s | 115,200,044 bytes | About 115 MB |
| C (10 min, track 1) | M4A, unchanged copy | 1.3 s | 12,263,586 bytes | About 12 MB |
| D (10 min) | MP3 | 9.6 s | 14,401,320 bytes | About 14 MB |
| D | M4A, AAC encode | 3.9 s | 14,518,706 bytes | About 14 MB |
| F (30 s, 5.1) | MP3, folded to stereo | 0.3 s | 721,320 bytes | About 721 KB |
| F | WAV, six channels kept | 0.3 s | 17,292,140 bytes | About 17 MB |
Every estimate landed within 2 percent of the saved file. On this machine MP3 ran at roughly 60 times real time, so an hour of audio took about a minute. The unchanged M4A copy was about ten times faster again, because nothing is decoded or encoded. WAV sat between the two. A phone will be slower; these are desktop figures.
What ffprobe found in the outputs
- Every MP3 was 48 kHz stereo at 192 kbps, as the size guide describes. MP3 durations ran 48 to 72 ms longer than the source, which is encoder padding at the end.
- Every unchanged M4A copy kept the source bitrate (128 or 160 kbps) and ran 21.3 ms longer than the source. That is the 1024 samples of AAC encoder priming that MP4 and MOV hide with an edit list; the copy keeps them at the start rather than altering any packet.
- WAV was 16-bit PCM at the source rate and channel count, including all six channels of file F.
- The two tracks of file C exported separately. The Microphone track is Opus, so an M4A from it is a new AAC encode, and the panel says so before Convert is pressed.
Two cases the test checks
An unchanged M4A copy from an MP4 or MOV keeps the 1,024 samples of AAC encoder priming at the start, which the source hides with an edit list. The first packet therefore has a negative timestamp (-0.0213 s at 48 kHz), and the copy shifts every packet by that offset without changing its bytes. That is why each copy above runs 21.3 ms longer than its source.
A 5.1 track cannot go into MP3, which holds at most two channels. MP3 export folds it to stereo and says so before Convert is pressed; WAV keeps all six channels, as file F shows.
Reading these numbers for your own recording
For an hour-long lecture or meeting saved as MP4 or MOV with AAC audio, the unchanged M4A copy is the quickest and smallest: about 72 MB at 160 kbps, with no generation loss. MP3 at 192 kbps costs 86.4 MB per hour and plays almost everywhere. WAV costs 691 MB per hour for 48 kHz stereo and is only worth it when the next step is an audio editor. If the File check shows a recording with no audio track, the sound was not captured; no converter can recover it. For opening problems, the troubleshooting section of the format guide lists what each message means.