Codec history · audio
Was this audio compressed, and at what quality?
To save space, a lossy encoder like MP3 or AAC quietly deletes the highest frequencies, and the lower the bitrate the more it cuts. That leaves a sharp horizontal edge in the sound's frequency content, and where that edge sits tells you roughly what quality the file was really encoded at, even if the file was later re-saved as something that looks higher quality. This finds the edge and reads it out. Read it as a strong lead, not a codec name.
Runs entirely in your browser. Your file is never uploaded or stored.
Decoding and analysing the spectrum…
The curve is the average energy at each frequency, low notes on the left, high on the right. A file that keeps energy all the way to the right edge is full-bandwidth. A lossy file drops off a cliff partway across and stays flat and silent above it: that cliff is the encoder's low-pass, and the further left it sits, the lower the bitrate the audio was really encoded at.
What the cutoff tells you
| Where the cutoff sits | Roughly what it means |
|---|---|
| At the very top (near 20-22 kHz) | Full-bandwidth: consistent with lossless (WAV/FLAC) or a top-tier encode (320 kbps / high AAC) |
| Around 19 kHz | Typical of a high-bitrate lossy encode (about 256 kbps) |
| Around 16-18 kHz | Typical of a mid-bitrate encode (about 128-192 kbps MP3) |
| Below about 15 kHz | A low-bitrate or heavily compressed source (or genuinely band-limited audio, like a phone recording) |
How it works
Lossy compression is built on a model of hearing: it spends bits where the ear is sensitive and starves the parts it can drop unnoticed, and the highest frequencies are the first to go. As D'Alessandro and Shi showed, the amount of energy left in the 16 to 22 kHz band is a reliable clue to the bitrate an MP3 was encoded at, with no need to see the file's headers. The tool averages the spectrum across the whole recording to get a stable picture, then walks down from the top to find where real energy begins, which is the cutoff. Kim and Rafii showed the same family of spectral traces can even identify which lossy format was used, though naming the exact codec needs the compressed bitstream, not just the decoded sound, so this tool stops at the cutoff and the bitrate band it implies.
Fake quality
The most useful case is a mismatch. If a file is delivered as a lossless WAV or FLAC, or labelled at a high bitrate, but its spectrum drops off a cliff at 16 kHz, then the audio was compressed hard at some earlier point and later wrapped in a higher-quality container. Yang, Shi and Huang called this fake-quality audio, and the cutoff exposes it: you cannot add back detail an encoder already threw away, so the low cliff survives the re-save. This tool flags that mismatch when it sees a lossless or high-rate container over a clearly lossy cutoff.
What fools it
A genuinely band-limited recording, an old tape, a telephone call, a voice memo, has little high-frequency content to begin with, so it can look like heavy compression even if it was never encoded. Re-encoding a lossy file again does not raise the original cutoff, so the tool reports the lowest bitrate in the chain, not the last one. And it reads the cutoff, not the codec, so it cannot tell MP3 from AAC. Pair it with the spectrogram to see the cutoff over time, and the audio metadata tool to compare what the file claims about itself.
Sources
- D'Alessandro, B. & Shi, Y. Q. (2009). MP3 Bit Rate Quality Detection through Frequency Spectrum Analysis. ACM Workshop on Multimedia and Security (MM&Sec) 2009:57-62.
- Kim, B. & Rafii, Z. (2018). Lossy Audio Compression Identification. 26th European Signal Processing Conference (EUSIPCO) 2018:2459-2463.
- Yang, R., Shi, Y. Q. & Huang, J. (2009). Defeating Fake-Quality MP3. ACM Workshop on Multimedia and Security (MM&Sec) 2009:117-124.