Everyone posts a speed-test screenshot, but the number that matters depends on the job. A twelve-stem pack takes about 11 minutes to upload at 10 Mbps and about 55 minutes at 2 Mbps. Neither figure says whether you can jam in real time. You can work out the answer for each job with arithmetic and two short tests.
One caveat first. The forum thread titled "What is your internet speed?" on the Internet for Musician board (audiosex.pro) shows no speed figures in its visible text, only related thread titles. One of those asks how to stream ASIO over the internet. So the numbers below are arithmetic from audio formats, not claims about what other people's connections achieve.
Does upload speed, download speed or latency matter most?
It depends on the job. For sending files, your upload speed sets the pace, because you are the one pushing data out. For receiving a collaborator's files or listening to their stream, your download speed matters, and most home plans have plenty of it. For real-time playing, neither headline number decides the outcome. Latency (how long a packet takes to travel) and jitter (how much that time varies) do.
Ask three separate questions:
- Is my upload fast enough to send files in a tolerable time?
- Is my latency low enough to play together?
- Is my connection stable enough that neither answer changes minute to minute?
A fast line on flaky Wi-Fi can fail the third question while posting a great screenshot.
This is an inference from how the pieces work, not a measured claim. A file transfer tolerates delay and jitter because the receiving end reassembles the data later, while live audio cannot wait. A slow upload therefore only costs you time on a transfer, but jitter can wreck a live session.
How long will my stems and mixes take to upload?
Start with the bitrate math. Uncompressed stereo at 24-bit/48 kHz is 48,000 samples × 24 bits × 2 channels = 2,304,000 bits per second, or 2.304 Mbps. That is 17.28 MB per minute (decimal megabytes). A mono stem is half that.
MP3 at 320 kbps is 0.32 Mbps, or 2.4 MB per minute. At 192 kbps it is 1.44 MB per minute, and at 128 kbps it is 0.96 MB per minute. FLAC is lossless, so its size depends on the material. Bounce your own file and read the size from disk rather than trusting a rule of thumb.
The formula is: seconds = file size in MB × 8 ÷ upload speed in Mbps. Run a speed test, note the upload figure, and fill in the last column of the worksheet. The example column assumes 10 Mbps purely for illustration.
| Format (stereo) | MB per minute | 3-minute mix | Example upload at 10 Mbps | Your upload time | |---|---|---|---|---| | 24-bit/48 kHz WAV | 17.28 | 51.84 MB | about 41.5 s | ______ | | FLAC | measure your bounce | measure your bounce | size × 8 ÷ 10 | ______ | | MP3 320 kbps | 2.4 | 7.2 MB | about 5.8 s | ______ | | MP3 192 kbps | 1.44 | 4.32 MB | about 3.5 s | ______ | | MP3 128 kbps | 0.96 | 2.88 MB | about 2.3 s | ______ |
These are ideal figures. Protocol overhead and other devices on your network add time, so multiplying by 1.25 gives a cautious planning estimate (an assumption, not a measurement). We cover related ground in see also: mpc hardware vs software: a level-matched sound test.
Now scale up to a stem pack. Twelve stereo stems of four minutes each at 24-bit/48 kHz come to 12 × 69.12 MB = 829.44 MB, or about 6,635 megabits. At 10 Mbps that takes roughly 11 minutes. At 2 Mbps it takes closer to 55 minutes.
Keep stems lossless regardless. The levers you control are when you upload (overnight works), whether you trim silence, and whether you send mono where the source is mono.
What is the lowest bitrate you can send for mix feedback?
No article can answer that for your ears, monitors and material, so run the test yourself. Do not trust anyone's claimed result, including a screenshot of a forum poll. The exercise takes about half an hour.
Set up the bounces
- Set your project to 48 kHz and 24-bit.
- Pick a dense 60-second section with hi-hats or cymbals, wide stereo elements and sharp transients, and loop it.
- Bounce it as WAV, FLAC, MP3 320, MP3 192 and MP3 128.
- Leave headroom on the master, with true peaks below about -1 dBTP, because lossy encoding can overshoot the original peaks.
Align and level-match
Import every bounce onto its own track and align them by sample, since MP3 encoders can add padding at the start. Then measure the integrated loudness of each bounce with a loudness meter. Trim gain until every file sits within 0.1 dB (or 0.1 LU) of the WAV. Louder tends to sound better, so skipping this step invalidates the test.
Run the blind comparison
Hide the filenames. Have a friend assign random letters, or use an ABX tool that randomizes for you. Run at least 10 trials per comparison. Scoring 9 or more out of 10 by chance happens about 1 time in 100, so that result suggests you really hear a difference. A score near 5 of 10 means you were guessing. We cover related ground in go deeper on how to test daws before buying: a 60-minute trial plan.
What to listen for
- Cymbal and hi-hat smear: does the sizzle turn into a smeared wash?
- Stereo width: do wide elements collapse toward the center or lose their edges?
- Transient softening: do snare and kick attacks lose snap?
Compare 128 against WAV first. If you can reliably pick it, step up to 192, then 320. Your lowest passing setting is your feedback format. Treat it as valid only for this mix style and your listening chain.
One rule holds regardless of your results: send lossless to anyone who will mix, master or process the audio further. Re-encoding a lossy file stacks artifacts.
What buffer size and setup should I start with for real-time sessions?
Start with a wired Ethernet connection, a 48 kHz session and a 128-sample buffer, then adjust by feel. Buffer delay is samples ÷ sample rate. At 48 kHz, 64 samples is about 1.33 ms, 128 is about 2.67 ms, 256 is about 5.33 ms and 512 is about 10.67 ms. Input and output buffers both count, so double the figure for a round trip through your interface.
Loopback test
Also read: see srv-330 reverb emulation vs stock reverb: a/b test guide
Measure your real figure by routing an interface output back into an input, using a cable or your interface's loopback feature. Record a click on one track and the returned signal on another. Zoom in and count the samples between them, then divide by 48,000 for seconds. Repeat at 64, 128 and 256 samples. If the audio crackles at 64, step up.
Network test
Ask your collaborator for their address, or the address of the session service, and run a one-minute ping. On Windows, use ping -n 60 followed by the address. On macOS or Linux, use ping -c 60. Read the minimum, average and maximum times, and note any lost packets. A big gap between minimum and maximum is jitter, and it matters more than a nice average.
There is no universal pass mark to quote here. As an inference, slow, sparse material forgives more delay than a fast groove, so play a click-led test together and trust your hands. The forum thread lists a question about streaming ASIO audio over the internet, which shows at least one musician is asking about it. The thread's visible text gives no setup you can rely on.
What should I check when things go wrong?
Start with the cable. Wi-Fi shares airtime and tends to produce the spikes that jitter checks reveal, so plug in with Ethernet before changing anything else. Then pause cloud backups, sample-library syncs, system updates and anyone else's streaming or uploads. A background upload can saturate your upload channel and push latency up for everything else.
Run the one-minute ping with those activities running and again with them paused. If the maximum time drops sharply, you have found your culprit. If it doesn't, restart the router and try a different time of day. Ask your collaborator to run the same test from their end too, because the problem may sit on their side.
Doesn't a faster speed test mean better remote jamming?
No. A speed test measures how much data you can move, not how quickly one small packet arrives or how consistently. An uncompressed 24-bit/48 kHz stereo stream needs only about 2.3 Mbps each way, so most home plans have the raw capacity. Delay and variation are what break sessions.
So stop comparing screenshots. Measure your upload for transfers, run the ping and loopback tests for live playing, and let a level-matched blind test decide how small your mix files can be.
