# Audio Spectrum Analyzer

> Upload audio and see its spectrum, peak frequencies, centroid, RMS/peak and waveform. Analyses an audio file (WAV directly; MP3, FLAC, Ogg, M4A and AIFF are decoded with ffmpeg first): averaged FFT spectrum chart, the strongest peak frequencies, spectral centroid and roll-off, spectral slope, RMS, peak, crest factor, DC offset, clipped samples, octave-band levels and a waveform overview.

Part of usefulweb.app (https://usefulweb.app/). Part of [Soundscape Atlas](https://soundscape.usefulweb.app/) (ownerApp: soundscape). Results are deterministic calculations from the inputs and the listed assumptions; the API returns exactly what the web page shows.

## Use it
- Human interface: https://soundscapespectrumanalyzer.usefulweb.app/ (interactive tool at https://soundscapespectrumanalyzer.usefulweb.app/tool)
- JSON API: POST https://soundscapespectrumanalyzer.usefulweb.app/api/v1/calculate with a JSON object of inputs (schema: https://soundscapespectrumanalyzer.usefulweb.app/api/v1/tool)
- OpenAPI: https://soundscapespectrumanalyzer.usefulweb.app/openapi.json
- MCP (Streamable HTTP): https://soundscapespectrumanalyzer.usefulweb.app/mcp - tool `soundscape_spectrum_analyzer`

## Inputs
- file (file, required): Audio file - WAV (PCM 8/16/24/32-bit or float), or MP3, FLAC, Ogg/Opus, M4A, AIFF; max 50 MB, the first 10 minutes are analysed. Files are processed in a private temporary folder and not kept.
- fftSize (select): FFT size - Larger = finer frequency resolution (sample rate ÷ size), fewer averaged frames.
- window (select): Window
- peaks (integer): Peaks to list

## Method
- WAV files are read by the engine (PCM 8/16/24/32-bit, float 32/64-bit, WAVE_FORMAT_EXTENSIBLE). Other formats are detected by their content and decoded by ffmpeg to 32-bit float WAV (the first audio track, up to stereo) in a private temporary folder.
- The channels are mixed to mono. The spectrum is a Welch average: frames of the chosen FFT size with 50 % overlap, each windowed and transformed with a radix-2 FFT; powers are averaged. Levels are corrected for the window’s coherent gain, so a sine of amplitude A reads 20·log10(A) dBFS at its peak.
- Peaks are local maxima that stand out from the median of the surrounding ±1/6 octave by more than random scatter of an averaged noise spectrum would allow (at least 9 dB; more when few frames carry the sound). Frequency and level are refined by a parabola through the three highest bins, so they are accurate to a fraction of a bin.
- Centroid = power-weighted mean frequency; roll-off = frequency below which 85 % of the power lies; slope = least-squares fit of 1/3-octave band levels from 100 Hz to 10 kHz (white noise ≈ 0, pink ≈ −3, brown ≈ −6 dB/octave). RMS, peak, crest factor and DC offset are computed on all samples.
- No loudness (LUFS) measurement: use the Audio Loudness Normalizer for that.

## Limits
- Peak frequencies are the strongest spectral lines, not necessarily the perceived pitch (a missing fundamental or a chord can sound different).