Audio Spectrum Analyzer & Spectrogram

About the Audio Spectrum Analyzer & Spectrogram

Look inside a sound. The spectrum shows how loud each frequency is and the spectrogram how that changes over time, for a file (MP3, M4A, WAV, FLAC, OGG or the sound of a video, up to 3 hours) or your microphone, live. Choose the FFT size and window, a logarithmic, linear or mel frequency axis, the level range and the colours, and read any point with the cursor: its frequency, the nearest musical note and its level in dBFS.

For a file the page also checks where the highest frequencies stop (a sharp cut-off is what lossy encoders leave, so a “lossless” FLAC or WAV made from an MP3 shows it), looks for mains hum at 50 or 60 Hz and its harmonics, and lists the strongest frequencies with their notes, beside the peak and RMS levels. Everything runs in your browser; nothing is uploaded or recorded. Without a pass the result is a free preview: the charts carry a MySmartCoPilot watermark and the list of strongest frequencies shows its first half.

Reading the charts

Levels are in dBFS, decibels below the loudest a digital file can hold: a full-scale sine wave reads 0 dBFS and everything quieter is negative. In the spectrum, the filled curve is the average level of each frequency over the whole file (live: now), the thin line the highest level each frequency has reached, and the dark line the moment at the playhead or under the pointer.

FFT size trades frequency detail for time detail: at 44.1 kHz, 4,096 points give 10.8 Hz per bin and a 93 ms window, 32,768 points 1.35 Hz per bin and 743 ms. Windows: Hann suits most sounds; Blackman and Blackman–Nuttall let quiet tones show next to loud ones, with wider peaks; Flat top reads the level of a tone most exactly; Rectangular gives the narrowest peaks but leaks the most. The logarithmic axis matches how we hear pitch, the linear one spaces harmonics and the top octave evenly, and the mel axis follows the perceptual scale used in speech analysis, m = 2595·log10(1 + f/700).

The frequency range check

Lossy encoders save space by cutting the highest frequencies away above a fixed point. The LAME MP3 encoder, for example, low-passes constant-bitrate MP3s at 17 kHz for 128 kbps, 18.6 kHz for 192 kbps and 20.5 kHz for 320 kbps by default. A file converted back to WAV or FLAC keeps that cut, so a “lossless” file with a wall in its spectrum was most likely made from a lossy one.

The check reads a high-resolution average of the file (under 3 Hz per bin at 44.1 and 48 kHz) and reports a sharp cut-off where the level falls by 20 dB or more within 500 Hz between 10 kHz and the top of the range with only noise above it; a gradual fade where the level is 30 dB below the middle frequencies (1 to 4 kHz) well before the top; otherwise content up to the top.

Where the wall stands matters, because sound at 44.1 or 48 kHz has a wall of its own: the filter that keeps it below half the sample rate, which at 44.1 kHz starts at about 20 kHz.

• below 19.5 kHz, the page reads it as a lossy encoder’s cut-off;
• from 19.5 to 21.5 kHz in a 44.1 or 48 kHz file, a lossy file at its highest bitrates and a genuine CD-quality recording look alike, and the page says it cannot tell;
• above that in a 44.1 or 48 kHz file, it is the end of CD-quality sound and no sign of a lossy file;
• in a high-resolution file (88.2 kHz and up), a wall at 24.5 kHz or below means the sound was most likely made at 44.1 or 48 kHz, or from a lossy file, and converted up; a wall higher still is the recording’s own filter.

It is evidence, not proof: some genuine recordings are filtered lower (old tapes, some masters), and lossy files at high bitrates can keep content to 20 kHz or more.

The hum check

Mains electricity runs at 50 Hz or 60 Hz, depending on the country, and a recording that picks it up — a ground loop, a long unshielded cable, a charger — shows a peak there and at its multiples. The page reports hum when the first or second harmonic stands at least 10 dB above the sound around it and at least two of the first four stand out by 6 dB or more, and names the harmonics it found. Each peak must lie within 0.3 Hz per harmonic of exactly 50 or 60 Hz and their multiples (50 ± 0.3 Hz, 100 ± 0.6 Hz …), so a bass note near them is not taken for hum: G1 is 49 Hz and A♯1 58.3 Hz.

Sources

• Window function: the window coefficients
• Mel scale: m = 2595·log10(1 + f/700)
• LAME MP3 encoder: its default low-pass for each bitrate (the freq_map table of libmp3lame/lame.c)
• Viridis, magma and inferno colour maps: released under CC0

Limitations

• Levels are digital (dBFS), not sound pressure: a phone or laptop microphone is not calibrated. For sound levels in dB(A), use the Decibel Meter.
• Files up to 3 hours. The spectrogram keeps at most 1,600 moments, so a long file shows less time detail; analyse a cut part for more.
• The microphone gives what the browser hands over: some phones and browsers keep processing the sound (noise suppression, automatic gain) even when asked not to.
• The checks are evidence, not proof: a filtered genuine recording can look lossy, and a lossy file at a high bitrate can look full.
• Old formats such as WMA and RealAudio cannot be read in the browser.

Privacy

On this device: nothing is sent

Everything happens in your browser. What you enter or open here is not uploaded or stored by MySmartCoPilot.

Open-source components

This page can load these open-source components, each only when a task needs it:

• It uses FFmpeg under the LGPLv2.1 and the LAME MP3 encoder under the LGPL, through libav.js: licence and source code.
• It uses FFmpeg under the LGPLv2.1, through the Mediabunny add-ons: licence and source code.
• It uses the LAME MP3 encoder under the LGPL, through the Mediabunny MP3 add-on: licence and source code.
• It uses Mediabunny under the MPL 2.0: licence and source code.

You may replace them with your own builds. Their licences and the credits of every library the site uses are on the open-source credits page.