Fake lossless detector: is your FLAC real?

Is that FLAC what it claims to be? This fake lossless detector measures how high a file's content reaches and tells you whether it is lossless or came from an MP3, from an upsampled "hi-res" release or from 16-bit audio dressed up as 24-bit. Load one or more FLAC, WAV, MP3, M4A or OGG files and they are analyzed in your browser, so nothing leaves your device.

The key test is the frequency cutoff. A 128 kbps MP3 made with LAME 3.100, the most widely used MP3 encoder, is lowpassed at 17 kHz and the detector measures it at 16.7 kHz, while a CD-quality FLAC reaches 22 kHz. It also looks for two tricks the cutoff can't reveal, the 96 or 192 kHz file that was made from a CD and the 24-bit file whose samples only use 16 bits.

How to use the fake lossless detector, step by step

Each track takes a few seconds, and the result fits on one line with the verdict, the cutoff, the sample rate and the bit depth.

1. Pick one or more files. With several, the detector works through them in a queue and ends with a summary table, handy when you want to check FLAC quality across a whole album.
2. Read the verdict and the cutoff in kHz, which is the number that decides.
3. Look at the spectrogram, with time running left to right and frequency bottom to top. An MP3 leaves a flat black band above its cutoff for the whole track, while real music fades out unevenly.
4. Check sample rate and bit depth. "24-bit (16 effective)" flags a 16-bit file padded with zeros, and "fake hi-res" means nothing is left above what a 44.1 or 48 kHz original would hold.
5. Hit "Copy verdict" and paste the result into a forum post or a complaint to the store.

What does each verdict mean?

The verdict comes from the measured cutoff. Above 20.5 kHz the spectrum is full, from 19 to 20.5 kHz it is almost full, from 16.5 to 19 kHz there is a mid cutoff and below 16.5 kHz a low one. Fake hi-res is judged separately, by looking at the ultrasonic band of files at 88.2 kHz or higher.

What cutoff each codec leaves, measured with this detector

A 320 kbps MP3 stops at about 20 kHz and a 128 kbps one at 16.7 kHz, but an MP3 V0 and a high-rate Vorbis leave no cutoff at all. We checked by encoding the same full-band test signal (pink and white noise at 44.1 kHz and 16-bit) with LAME 3.100, libvorbis and libopus, then running every version through the detector itself. With real music the cutoffs usually come out a little lower, because music carries less energy up there than noise does.

AAC is missing because its cutoff depends on the encoder (Apple, FDK or FFmpeg) and we haven't measured it yet.

The rule forums and AI summaries repeat, that 128 kbps cuts off at 16 kHz, doesn't describe LAME well, since its own source code sets 17 kHz at that bitrate and 20.5 kHz at 320. A 320 kbps MP3, Opus and Vorbis q5 land in the same gray zone as a legitimate master filtered near 20 kHz, and the "almost full" verdict says exactly that. And because LAME 3.100's V0 applies no lowpass and libvorbis stops filtering from q6 upward, no cutoff-based tool can catch them. If you want to choose a bitrate, our guide to what bitrate you need for music puts numbers on it.

Five times it pays to check whether a FLAC is real

It pays when a file comes from a shady source, when you paid for high resolution, when you inherit an old library, before a DJ set and before a blind test.

A download that claims to be FLAC

If a FLAC comes from a peer-to-peer network, a forum or an out-of-print album someone digitized years ago, run it through the detector before you file it away. Converting an MP3 to FLAC recovers nothing, because the file grows to FLAC size while the MP3's lowpass stays baked in. A fake FLAC made from a 128 kbps MP3 shows up here with a cutoff near 16.7 kHz and a "mid cutoff" verdict, and one made from 320 kbps lands around 20 kHz. Analyze the whole album rather than a single track. If several songs come back with a mid cutoff, the source is unreliable even if a couple look fine, because whoever assembled it mixed sources. The reverse logic holds too, with a caveat. A 22 kHz cutoff proves nothing on its own, since a very high-bitrate MP3 reaches it as well, but it does rule out the crudest case, the cheap MP3 converted and relabeled. And since the detector runs in any recent browser, phone included, you can check a download the moment it lands.

Before you complain about a hi-res purchase

If you paid for a 24-bit/96 kHz download, check it before you accept the purchase, because it costs more than the CD version. The detector decodes the file at its native rate and looks at the ultrasonic band. If the music reaches 20 kHz, nothing is left above roughly 21-26 kHz and the drop has the shape of a resampling filter, the verdict is "fake hi-res", which points to a 44.1 or 48 kHz master converted afterwards. Treat it as a strong hint worth double-checking, because a genuine hi-res recording whose music ends there (because of the microphones or a lowpass filter somewhere in the recording chain) leaves a similar trace. The detector also measures how many of those bits the samples use, and that test is more clear-cut. A file that claims 24-bit but whose lowest eight bits are always zero shows up as "24-bit (16 effective)". When we checked three tracks from an album circulating as 24-bit/88.2 kHz, all three came out that way, and FFmpeg confirmed it independently. If it happens to you, ask the store for an explanation and attach the verdict and a screenshot of the spectrogram. And if you are not sure hi-res is worth it for you, our guide on who needs lossless audio will help you decide.

Auditing an old library in batches

For a collection that has lived on several hard drives, select every track of an album or a few dozen files at once. The detector processes them one by one and returns a table with verdict, cutoff, sample rate and bit depth for each. The odd ones stand out immediately, because a single track cutting off at 16.7 kHz on an album where the rest reach 22 kHz is almost always an MP3 someone used to fill a gap. That consistency test is more reliable than judging one file in isolation, since the rest of the album shows you what that recording normally looks like. Everything happens in your browser, so there is no upload quota and no waiting in line. Memory is the limit, because a 20-minute track at 192 kHz takes up almost 2 GB once decoded, so on a phone it may not fit, and it's better to analyze it on a computer. Save the copied table next to the album folder, and the report will be there if a track ever raises doubts again.

If you DJ, check the pack before the gig

Run the promo folder through the detector before the gig, because packs, compilations and "remasters" passed between DJs are fertile ground for transcodes. A track that cuts off at 16 kHz can go unnoticed on headphones, yet next to a clean track, on the same system and at the same level, it sounds dull, and it is better to find out at home than in the booth. Copy the summary table with one click and set aside anything marked mid or low cutoff. If a track you need only exists in that quality, at least you will know what to expect, since boosting the treble won't bring back what the lowpass removed and only lifts the noise and harshness that remain. For the "almost full" gray zone, compare against another copy of the same release before you discard it. When a pack sold as lossless arrives with several tracks cut at 16 kHz, the problem lies with whoever prepared it, so let them know before the bad copies spread to other DJs.

Before an ABX test or a comparison of two editions

Before you compare a FLAC with an MP3 blind, or two editions of the same album, make sure each file is what it claims to be. A "FLAC" that turns out to be a 320 kbps MP3 turns the test into MP3 versus MP3, and any conclusion comes out skewed. Run both files through the detector and note cutoff and bit depth. If they match suspiciously, find another source. With verified files you can move on to the question that matters, which our guide on whether you can hear the difference between FLAC and MP3 digs into. Hearing matters too, because sensitivity above roughly 16 kHz drops with age. Check the highest frequency you can hear before worrying about a cutoff at 19 kHz, since a limit your ears don't reach can't spoil the comparison.

When the detector gets it wrong

The detector fails in three ways. It flags some legitimate files that simply have no treble, it lets very high-bitrate lossy files through and it can mistake a genuine hi-res recording for an upsampled one. It also has two limits worth knowing.

• False positives. Old or narrow-band recordings (shellac, cassette, plenty of tape from the fifties and sixties), solo voice or solo piano may have no energy above 15-18 kHz. On the spectrogram they give themselves away because the energy fades gradually and unevenly, without a codec's flat shelf. It doesn't always happen. Two real 24-bit/96 kHz transfers of shellac records we tested came out as full spectrum, because their surface noise reaches 30 kHz, so the false positive shows up mostly when a transfer has been filtered or denoised.
• False negatives. LAME 3.100's V0 and Vorbis from q6 (around 192 kbps) apply no lowpass, so they reach 22 kHz and the detector calls them full spectrum, as would any tool that relies on the cutoff. And an upsampled hi-res file whose music doesn't reach 20 kHz gets a low-cutoff verdict instead of fake hi-res, because the cutoff takes priority.
• Genuine hi-res with no ultrasonic content. A real 96 kHz file whose music dies out around 20-25 kHz, because of the microphones or a lowpass filter in the recording chain, can come out as fake hi-res. That is why the verdict is a hint worth checking against the CD version.
• Browser limits. To check fake hi-res and bit depth, the file is decoded at its native rate. If your browser can't do that, the detector says so and sticks to the cutoff. Bit depth is only measured in lossless formats such as WAV and FLAC.
• A valid FLAC can still be fake. flac -t checks that a file isn't damaged and that its MD5 signature matches the audio, but a transcode passes it, because the signature was computed on audio that was already degraded.

The spectrum tells you what the file contains, not whether you can actually hear the difference. For that second question there is the blind ABX test.

Frequently asked questions