# Spectrogram & audio analyzer > How do you find hidden sounds in a recording? Open a sound or video file or record one, see it as a spectrogram, bring out faint details and listen to them isolated. Free spectrogram and audio inspector: open audio or video, or record, then zoom, recolour, isolate and play faint sounds. Runs in your browser. - Page: https://how.box/spectrogram/ - Short address: https://how.box/audio-analyzer/ - Runs entirely in the browser. - License: HowToolbox's own text and data (titles, explanations, FAQs, tool descriptions and schemas, this file) are CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/); AI training and answers are welcome. Attribute "HowToolbox (how.box)" with a link to the page used. Third-party data keeps its own terms; the name and logo are not licensed. Publisher: HowToolbox, contact@how.box. Terms: https://how.box/about/terms/ ## Inputs Answer link: https://how.box/spectrogram/?c={c}&inv={inv}&lo={lo}&hi={hi}&b={b}&k={k}&g={g}&auto={auto}&dn={dn}&tl={tl}&s={s}&fl={fl}&fh={fh}&n={n}&w={w}&o={o}&md={md}&temp={temp}&tu={tu}&su={su}&hr={hr}&hf={hf}&hs={hs}&st={st}&pi={pi}&pr={pr}&pl={pl}&pd={pd}&ts={ts}&tr={tr} - `c` (magma, inferno, viridis, grey or hot): Colour map. Example: `viridis`. - `inv` (boolean): Reversed colour map. Example: `true`. - `lo` (number -160 to 60): Darkest level, dB. Example: `-90`. - `hi` (number -100 to 100): Brightest level, dB. Example: `-30`. - `b` (number -50 to 50): Brightness shift, percent. Example: `10`. - `k` (number 25 to 400): Contrast, percent. Example: `150`. - `g` (number 0.3 to 3): Gamma of the colour scale. Example: `0.7`. - `auto` (boolean): Auto levels for the view. Example: `false`. - `dn` (boolean): Noise-floor removal. Example: `true`. - `tl` (number 0 to 12): Treble lift, dB per octave. Example: `3`. - `s` (linear, log or mel): Frequency axis. Example: `linear`. - `fl` (number 0 to 190000): Lowest frequency shown, Hz. Example: `100`. - `fh` (number 100 to 192000): Highest frequency shown, Hz. Example: `8000`. - `n` (256, 512, 1024, 2048, 4096, 8192 or 16384): FFT size, samples. Example: `8192`. - `w` (hann, hamming, blackman-harris or rectangular): Window function. Example: `blackman-harris`. - `o` (50, 75 or 87.5): Window overlap, percent. Example: `87.5`. - `md` (air, fresh-water or seawater): Medium of a traced moving source. Example: `seawater`. - `temp` (number -60 to 140): Air temperature in tu. Example: `5`. - `tu` (c or f): c or f. Example: `f`. - `su` (kmh, mph or ms): Unit of a traced source speed: kmh, mph or ms. Example: `ms`. - `hr` (auto, off, slower, faster or shift): Playback of sound outside hearing: auto, off, slower, faster or shift. Example: `slower`. - `hf` (auto, 2, 4, 8, 10, 16, 20, 32 or 64): Slower or faster factor. Example: `10`. - `hs` (number -190000 to 190000): Frequency shift, Hz (negative: down). Example: `-36000`. - `st` (sound or picture): What the empty page offers first: sound or picture. Example: `picture`. - `pi` (boolean): A placed picture: dark parts loud. Example: `true`. - `pr` (number 20 to 90): A placed picture: dB from white to black. Example: `60`. - `pl` (number -40 to 20): A mixed picture: level against the sound under it, dB. Example: `-12`. - `pd` (number -40 to 0): A mixed picture: the sound in its box lowered by dB. Example: `-20`. - `ts` (number 1 to 60): Empty track for a picture alone, seconds. Example: `10`. - `tr` (22050, 44100, 48000 or 96000): Sample rate of that empty track, Hz. Example: `48000`. Example answer: https://how.box/spectrogram/?c=viridis&inv=1&lo=-90&hi=-30&b=10&k=150&g=0.7&auto=0&dn=1&tl=3&s=linear&fl=100&fh=8000&n=8192&w=blackman-harris&o=87.5&md=seawater&temp=5&tu=f&su=ms&hr=slower&hf=10&hs=-36000&st=picture&pi=1&pr=60&pl=-12&pd=-20&ts=10&tr=48000 The page computes the answer in the browser: read results from the agent tools below, not from the page HTML. ## Agent tools - `spectrogram_summary`: Summarises the sound the visitor opened (an audio or video file) or recorded on this page: duration_s, sample_rate_hz (the analysis rate: the file's own for WAV and FLAC, up to 384 kHz), nyquist_hz (the highest frequency it holds), peak_dbfs (largest sample, dB below full scale), rms_dbfs (RMS loudness, AES17 dBFS: a full-scale sine is 0), clipped_percent, and strongest (the strongest frequencies of the average spectrum: frequency_hz, level_db, loudest first, at least a sixth of an octave apart, from 10 Hz), ultrasonic_percent and infrasonic_percent (share of power above 18 kHz and below 40 Hz), energy_range_hz ([low, high]: the 1st and 99th percentile of power over frequency, works for ultrasonic files). - MCP, one POST to https://how.box/mcp: `{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"spectrogram_summary","arguments":{}}}` - GET: https://how.box/mcp/run/spectrogram_summary ## How it works A spectrogram is a picture of sound: time runs from left to right, pitch (frequency) from bottom to top, and the brighter a spot, the louder that frequency was at that moment. A sound covered by a louder one often sits elsewhere on the picture, so the eye finds it before the ear: a distant aircraft as a slow curve, a bird above traffic, a hum as a steady line. The picture comes from the short-time Fourier transform. The sound is cut into short overlapping windows of N samples, and each is split into its frequencies: X(k) = Σ x(n)·w(n)·e^(−2πikn/N), where x is the sound, w the window that fades each piece in and out, and k a frequency step of f_s/N hertz (f_s is the sample rate). A longer window separates close tones better but blurs quick changes: at 48 000 samples a second, 2048 samples are 43 ms long and 23 Hz apart, 512 samples 11 ms and 94 Hz. Isolating keeps the frequencies inside your selection, sets the rest to zero and turns the pieces back into sound; slowing down keeps the pitch by moving each frequency's phase. A recording holds nothing above half the rate it was sampled at (the Nyquist limit: 24 kHz for most files), isolation leaves some faint ringing, and a sound that shares its time and frequency with louder noise cannot be pulled apart. Sources: J. O. Smith, Spectral Audio Signal Processing and Digital Audio Resampling (CCRMA, Stanford); W3C Web Audio API. A picture can go the other way: each row of it becomes a steady tone and each column a moment, as loud as the pixel is bright, so the spectrogram draws the picture. The rows follow the frequency scale you place it on (linear, log or mel), so it looks right on that scale and bends on another. Lossy formats such as MP3 or AAC throw away the faint details it is made of; save it as WAV. ## FAQ **Is my recording uploaded?** No. Files and microphone recordings are decoded and analysed in your browser, and nothing is sent or stored. A copied link holds only the display settings. **Can I open a video?** Yes. MP4 or MOV with AAC sound and WebM videos work in most browsers; only the sound track is used and the picture is ignored. If a format (such as some MKV files, AC-3 or HEVC video) cannot be decoded, convert it to MP4 with AAC sound, MP3 or WAV. **Can I see and hear bat calls?** Yes, from a recording made for them: most bats call above 20 kHz, and an ordinary recording stops at half its sample rate, usually 22 or 24 kHz, so you need a bat detector or recorder at 192 or 384 kHz (WAV or FLAC). This page shows everything up to half that rate and brings it into hearing: played ten times slower, a 40 kHz call is heard at 4 kHz and lasts ten times longer (time expansion), and played ten times faster an elephant's 15 Hz rumble becomes a 150 Hz hum. A shift instead moves every pitch by the same amount in real time, as a heterodyne bat detector does. A phone or ordinary microphone filters ultrasound out. **How do I measure the speed of a passing car, train or plane?** Record it passing, choose Trace a curve and follow one clear tone (an engine or tyre hum, a horn, or one of its harmonics) from well before to well after the moment its pitch falls fastest. The Doppler effect f = f₀·c/(c − v·cosθ) (f₀ its own pitch, c the speed of sound, v its speed, θ the angle between its path and the line to you) is fitted to the trace and gives the speed, its own pitch and when it was closest: a 400 Hz horn passing at 60 km/h in air (c = 343 m/s) is heard at about 420 Hz coming and 382 Hz going. Set the air temperature, or water for a boat heard under water. A short or wandering trace gives no speed or only a rough one. **The quiet sound is still buried. What helps?** Turn on Remove steady noise to fade hum and hiss, lower Gamma to lift faint levels, zoom in on the frequencies where the sound is, and try Narrow tones for a steady whistle or Sharp transients for clicks. A treble lift helps with high, weak sounds. **Why does the isolated sound ring or warble?** Cutting frequencies out leaves small traces at the edges of the selection, sometimes called musical noise. A slightly larger selection, a smaller window or the Original switch for comparison helps. It is a tool for finding and checking sounds, not for studio restoration. **Can I hide a picture in a sound?** Yes. Open the sound, choose a picture under Picture, drag and resize it to the time and pitch you want, set how loud it is against the sound (and lower the sound under it if it should read clearly), then Mix in and download the track as WAV. With no sound open, the picture becomes a sound of its own. More: [HowToolbox index](https://how.box/llms.txt) · [Agent Skill](https://how.box/skill.md) · [agent terms](https://how.box/about/agents/)