44.1 or 48 kHz, 16 or 24 bit: which should you pick, and what does converting cost?

What the two numbers are

The sample rate is how many times per second the sound is measured; the highest frequency a file can hold is half of it, so 22,050 Hz for 44.1 kHz and 24,000 Hz for 48 kHz. The bit depth is how finely each measurement is written down: 16-bit has 65,536 steps, 24-bit has 16,777,216, and 32-bit float writes each value as a decimal number that can also go past full scale. Neither number says anything about what is in the file. An MP3 opened on this site has nothing above its own cutoff (18.2 kHz at 192 kbps in our test) whatever rate you write it out at, and a 24-bit WAV of a 16-bit recording has eight bits of zeros.

So the real question is not which rate is better but what converting between them does, because every conversion goes through a filter. This page measures that filter.

How we tested

• Pages: a local copy of this site's WAV to MP3 page (output WAV 16-bit at another sample rate, and MP3) and MP3 to WAV page (16, 24, 32-bit float; 22.05 to 96 kHz; volume), in Chromium 153 driven by Playwright 1.63. The files are what the browser downloaded.
• Test signals, written by a script as 24-bit WAV: a 20-second sweep from 20 Hz to just under half the sample rate; single tones, 2 seconds each at -12 dBFS, from 1 kHz up to 23.5 kHz (at 48 kHz) and 40 kHz (at 96 kHz); and 65 seconds of 1 kHz plus 10 kHz for the seam test, since the WAV to MP3 page converts 30 seconds at a time.
• Music: one minute (1:00 to 2:00) of Mozart's Symphony No. 40, first movement, by the Musopen Symphony, public domain on Wikimedia Commons as a 48 kHz 24-bit FLAC. FFmpeg 4.4.2 with the soxr resampler made the 44.1 kHz and 48 kHz 24-bit WAVs, the floating-point references at each rate, and the same conversions for comparison. The MP3 inputs were made with this site's encoder (lamejs) at 192 kbps.
• Score is signal-to-error ratio in dB against the reference after aligning: how far the waveform moved, not what you hear. Tone levels are read from a spectrum of the middle 1.4 s of each tone and compared to the same reading of the source file.

What the sample rate converter does to the top of the band

A tone that stays well under the new half-rate comes through unchanged; a tone near it is turned down; a tone above it is removed, with a little of it folding back as a mirror image.

The sweeps put the corners at the same place in both directions: -1 dB near 18.3 kHz, -3 dB at 19.2 kHz, -6 dB at 20.0 kHz (19.9 kHz going down), -20 dB at 21.5 to 21.7 kHz. Below 18 kHz the sweep stayed within 0.8 dB of the source. Everything a 20-year-old can hear is on the flat part; the loss is the top 2 kHz of a 44.1 kHz file, turned down rather than removed.

From 96 kHz the corners move: to 48 kHz the filter is -3 dB at 20.1 kHz and -6 dB at 21.6 kHz, with a 26 kHz tone folding back to 22 kHz at -25.4 dB and a 30 kHz tone at -81 dB; to 44.1 kHz it is -3 dB at 18.4 kHz and -6 dB at 19.9 kHz, with 24 kHz folding to 20.1 kHz at -23.7 dB, 26 kHz to 18.1 kHz at -41.2 dB, and 30 and 40 kHz gone entirely (digital silence in the 16-bit file). Real recordings carry very little above 22 kHz, so this fold-back mostly matters for synthetic or ultrasonic material.

Music, length and seams

• The 60 to 64 dB is all in the top band. Comparing this site's 44.1 to 48 kHz file with FFmpeg's 44.1 to 48 kHz file of the same WAV, after cutting both at 18 kHz, gave 73.4 dB, which is the 16-bit floor from the first row; 89% of the difference between them was above 18 kHz. Going down (48 to 44.1 kHz) it was 72.8 dB below 18 kHz, 88% of the error above. FFmpeg's resampler keeps the band flat further out, which is why its 48 to 44.1 kHz score is 16 dB higher; below 18 kHz the two are the same.
• The 44.1 to 48 kHz rows score lower than the same-rate rows for a second reason: the 48 kHz reference still has the original's content between 20 and 24 kHz, which no 44.1 kHz file can carry. FFmpeg's version is 62.4 dB for the same reason. If your file is already 44.1 kHz, that content is already gone and the conversion costs you only the top-band filtering above.
• No seams. The 65-second two-tone file came out as exactly 3,120,000 samples at 48 kHz. Fitting the two tones over the whole file and looking at what was left gave -93.2 dBFS in every 100 ms block, including the blocks at 30 and 60 seconds where the page joins its pieces: the residue is the 16-bit rounding and nothing else. The output also started on the same sample as the input (no delay).
• Length: upward conversions were exact to the sample; 48 to 44.1 kHz came out one sample (0.02 ms) short, as we also saw in the compress WAV test.

Why the corner sits at 19 to 20 kHz

Both converter pages decode your file with the browser's Web Audio API at the output rate you chose. The Web Audio API 1.1 draft (W3C Working Draft, 22 September 2026, read 2026-10-26) tells the browser, in the decoding steps, to "resample it to the sample-rate of the BaseAudioContext if it is different from the sample-rate of audioData". How is up to the browser. In Chromium's source, read 2026-10-26, audio_bus.cc does this with a media::SincResampler asked for 64 frames at a time, which by sinc_resampler.h selects the smaller 32-point kernel (the file says the larger 64-point one is the aim "for perceptible audio quality", with 32 as the fallback). In sinc_resampler.cc, the low-pass cutoff for that kernel size is multiplied by 0.90, with the comment that the cutoff is moved "slightly downward to avoid" aliasing and that the values "are derived empirically". 0.90 times 22,050 Hz is 19,845 Hz, which is where we measured -6 dB. A short kernel also means a gradual slope, which is why a 22 kHz tone is only 25 dB down and a little of it mirrors back. Other browsers may resample differently; we only tested Chromium.

MP3 at 44.1 vs 48 kHz

The sample rate of an MP3 changes nothing you can measure here: the encoder cuts the top at 18.2 kHz (192 kbps) or 20 kHz (320 kbps) at either rate, and the score is the same to 0.1 dB. The 48 kHz file is a few hundred bytes smaller only because its frames are shorter and the padding at the end differs. Converting the 48 kHz WAV down to a 44.1 kHz MP3 cost nothing either (27.6 dB), because the MP3's own cutoff sits below the resampler's. Pick the rate your destination wants.

16, 24 or 32-bit float when writing a WAV

• The float output is the decoder's own numbers. It matched FFmpeg's decode of the same MP3 to 122.9 dB (largest difference 0.05 of a 16-bit step), so the 32-bit float WAV is as exact a copy of what is in the MP3 as you can get. The 24-bit file is within 1 step of it everywhere and rounds correctly (98% of samples within half a step). The 16-bit file is written by truncation rather than rounding, which is why its noise is 6 dB above the ideal and its largest error 1.7 steps; the noise is still at -94.7 dBFS, 46 dB below the MP3's own error, which measured -48.7 dBFS against the master.
• None of it changes the score against the master: 27.5, 27.6, 27.6 dB. What the MP3 lost (-48.7 dBFS) is 100 dB louder than what 24-bit saves over 16-bit. For a file you will play, hand in or burn, 16-bit is the right choice and half the size of float. For a file that will go through more processing (edits, effects, another export), 24-bit keeps the rounding out of the way for one more round.
• Volume above 100% is the one case for float. At 150% the float file had 30 samples past full scale (peak +2.12 dBFS, and the gain came out as 3.522 dB, exactly 1.5 times); the 16-bit and 24-bit files clipped those same 30 samples at full scale. At 100% the loudest sample was 0.851, so nothing clipped at any depth.
• Float does not save an MP3 that is itself too hot. A 128 kbps MP3 of a limited, 0 dBFS master decodes in FFmpeg to a peak of +1.84 dBFS with 2,499 samples over full scale. The page's 32-bit float file of it peaked at exactly 1.0, with those 2,499 samples sitting at 1.0: the browser's decoder clamps before the bit depth is applied, as the MP3 to WAV FAQ says. If you need those peaks, turn the volume down on this page; it is applied after the clamp, so it cannot bring them back either. See how much headroom an MP3 needs for the encoder side of this.
• Going the other way, 24-bit to 16-bit WAV on the WAV to MP3 page scored 73.5 dB against the 24-bit master at both rates; FFmpeg without dither scored 73.9 and 74.0 dB. Neither page adds dither.

Writing an MP3 out at a higher or lower rate

The MP3 to WAV page also lets you pick 22.05, 48, 88.2 or 96 kHz for the WAV. From the 44.1 kHz 192 kbps MP3:

Writing an MP3 out at 48 or 96 kHz adds samples and bytes (2.18 times the size at 96 kHz) and nothing else: the band above 20 kHz holds -81 to -105 dB of the energy, which is the resampler's leftovers, not music. Because the MP3's content ends at 18.2 kHz, under the converter's corner, the 48 kHz float file agreed with FFmpeg's resampling to 87.7 dB, far better than the 64 dB we got on the orchestral WAV that had content up to 22 kHz. The 22.05 kHz option is different in kind: it cuts everything above about 10 kHz (the converter's corner scales with the rate) and scored 48.2 dB against a careful resample of the same decode, so use it only when something demands that rate.

What to choose

1. Find out what the destination wants. If a program, device or upload form names a rate or bit depth, use it. If it doesn't, keep the file's own sample rate: every conversion runs the filter above once, and two conversions run it twice.
2. MP3 to WAV: open MP3 to WAV, leave Same as the MP3 and 16-bit PCM for a file you will play or submit; pick 24-bit PCM if the WAV goes into further editing; pick 32-bit float only if you also raise the volume above 100%. Pick 48 kHz only if the destination needs it; it costs nothing measurable with an MP3 source but a 9% larger file.
3. WAV to WAV at another rate or depth: open WAV to MP3, set Output format to WAV (16-bit PCM) and the sample rate you need. This path always writes 16-bit; this site cannot write a 24-bit WAV from a WAV, only from an MP3.
4. WAV to MP3: on the same page choose the bitrate first (that is what sets the quality, see how much each setting saves) and the sample rate the destination wants; 44.1 and 48 kHz gave the same score.
5. Keep the original. Every conversion here is a copy.

Questions

Should I convert my 44.1 kHz music to 48 kHz before putting it in a video?

If the video editor or the delivery spec asks for 48 kHz, yes, once, on the WAV to MP3 page with WAV output: it costs 0.8 dB at 18 kHz, 3 dB at 19.2 kHz and nothing below. If nothing asks, leave it; the editor will do its own conversion when it has to, and you avoid doing it twice.

Does 48 kHz or 24-bit make an MP3 sound better?

Not in anything we could measure. 192 kbps scored 27.5 dB at 44.1 and 48 kHz with the same 18 kHz cutoff, and the 24-bit source gave the same MP3 as the 16-bit one. The bitrate is the dial that matters for an MP3.

Can I get a 24-bit or 96 kHz WAV out of a WAV here?

No. The WAV to MP3 page writes 16-bit WAV at 16 to 48 kHz. 24-bit, 32-bit float, 88.2 and 96 kHz WAV output exists only on the MP3 to WAV page, from an MP3. We tested Chromium only; other browsers' converters may behave differently.