An interactive audio analysis laboratory that runs entirely in your browser.
Open a recording and measure it: no upload, no account, no server doing the work.

Features · Gallery · Quick start · How it works · Validation · Limitations · Technical reference

Not an audio player with charts

ECHOFORM treats a recording as something to investigate. Every number comes from the decoded samples, every algorithm is tested against signals with known answers, and every estimate says how it was made and when it should not be trusted. If a recording cannot support a measurement, the app says so instead of showing a plausible figure.

• Honest by design: no simulated charts, and undefined values stay undefined, never zero
• Private by default: audio is decoded and analysed in your browser tab and never uploaded
• Built for long files: analysis runs in a cancellable worker, so the page stays responsive
• Non-destructive: filters render into a separate buffer and the original is never changed

Features

Waveform	Multiresolution rendering, from whole recording down to individual samples
Spectrum	Calibrated FFT with 7 sizes and 4 windows, averaged or live at the playhead, with peak detection
Spectrogram	Time-frequency regions, band zoom, PNG export, and detail recomputed as you zoom
Signal analysis	Level timeline, histograms, silence regions, clipping intervals and DC offset
Spectral descriptors	Centroid, rolloff and flatness for any time selection or frequency band
Stereo analysis	Correlation over time, mid and side levels, a width index and a vectorscope
Filter Laboratory	7 filter types in a chain, exact response curves, and original versus processed listening
Comparison Lab	Two recordings with explicit alignment, normalisation, and spectral and waveform differences
Events and pitch	Onsets, tempo with alternatives, a pitch track, harmonics and harmonic-to-noise ratio
Noise analysis	Estimated floor and SNR with method and confidence, plus mains hum indicators
Reports	JSON, CSV and a standalone printable document, with configuration and methods included

Every measurement has an inline explanation with its formula and what it can and cannot tell you.

Gallery

Spectral Analysis Calibrated spectra, peaks and descriptors	Stereo Analysis Correlation, mid and side, and a vectorscope
Events and Pitch Onsets, tempo candidates and noise evidence	Filter Laboratory Exact response, measured result
Comparison Lab Aligned waveforms and overlaid spectra	Reports JSON, CSV and printable HTML

_{Designed for the desktop, and still usable on tablets and phones.}

_{All screenshots use a generated demo signal.}

Quick start

You need Node.js 20 or newer.

git clone https://github.com/OwXMaybe/echoform.git
cd echoform
npm install
npm run dev

Open http://localhost:3000, then drop in an audio file or choose Use test signal.

Command	What it does
npm run dev	Start the development server
npm run check	Run types, lint, unit tests and a production build
npm run test:e2e	Run the browser workflows in an installed Microsoft Edge

Keyboard

Key	Action	Key	Action
Space	Play or pause	[ ]	Set selection start or end
S	Stop	Esc	Clear the selection
L	Loop the selection	← →	Seek one second
M	Mute	+ -	Zoom
Home	Go to the start	0	Show the whole recording

How it works

flowchart LR
    F["Audio file"] --> D["Format check and decoder"]
    D --> B["AudioBuffer<br/>(playback)"]
    D --> W["Analysis worker"]
    W --> J["Cancellable DSP jobs<br/>FFT, STFT, filters, YIN, flux"]
    J --> S["External stores"]
    S --> C["Canvas charts and panels"]
    S --> R["Reports<br/>JSON, CSV, HTML"]
    B --> O["Audio output"]
    C -- "selection, zoom, settings" --> J

Loading

Signal processing lives in src/core/dsp as plain TypeScript with no React or DOM, so it can be tested directly. The FFT, windows, filters and every estimator are implemented in this repository. PCM data never enters React state, and charts paint to canvas outside the React render cycle, so moving the playhead does not repaint the data underneath it.

src/core/dsp             Signal processing: FFT, STFT, filters, pitch, onsets, tempo, noise
src/core/audio           Format identification, decoding, playback, test signals
src/core/analysis        Worker client and measurement definitions
src/core/visualization   Scales, colour maps and plot geometry
src/core/reports         Report model and its JSON, CSV and HTML forms
src/workers              Analysis worker and its typed protocol
src/stores               External state; PCM is never stored here
src/components, src/app  Interface
tests/unit               Scientific validation
tests/e2e                Browser workflows

Scientific validation

The unit tests generate signals with known properties and compare each result with the expected answer. Some of the checks:

Algorithm	Test signal	Verified result
Spectrum	Sine of amplitude A	Reads 20 log10(A), for every window and FFT size
RMS	Full-scale sine	−3.01 dBFS
Correlation	Identical, inverted and quadrature channels	+1, −1 and 0
Peaking filter	+12 dB at 1 kHz	Exactly +12 dB at the centre, with stable poles across all designs
Offset estimate	Noise cut 12,345 samples late	Recovered to the exact sample
Pitch	65 Hz to 1 kHz sines	Within 5 cents, with white noise correctly left blank
Tempo	Click tracks from 70 to 150 BPM	Within 1 BPM, with noise and irregular events giving no tempo
Noise floor	Tone bursts over known noise	Within 1 dB, withheld for continuous audio

Fourteen browser workflows then drive the production build through real file handling, playback, each lab, report downloads and settings storage. They also confirm the app makes no POST or third-party requests, run axe accessibility rules on every page, check phone-width layout, and verify that charts paint.

Performance: on one development PC, a 10-minute stereo 48 kHz file showed its waveform in 2.5 s and finished all background analysis in about 20 s, with a median zoom frame of 7.5 ms.

Privacy

Audio is decoded and analysed locally. There is no account, upload or analytics, and recordings are never saved: closing the recording or the tab releases them. Storing your preferences is off by default, and turning it off deletes them. Reports can leave file names out, and they stay on your device unless you share them.

Limitations

ECHOFORM states what it cannot do. The most important limits are:

• Tested on generated audio. All tests use synthetic WAV files. Onset, tempo, pitch and noise behaviour on real music and speech has not been measured, and MP3, FLAC, Ogg and M4A decoding has not been tested with real files.
• Digital levels only. Measurements are not LUFS, not true peak, and not calibrated to acoustic sound pressure.
• Estimates are estimates. Tempo can report half or double the true value, pitch needs a single periodic source, and the noise floor assumes the quietest passages are background.
• Clipping means digital clipping. Analog distortion cannot be detected this way.
• Browser coverage. Workflows ran on Microsoft Edge only, and accessibility was checked with automated rules rather than a screen reader.

The technical reference has the full list and every measurement convention.

Roadmap

• Validation on real music and speech, with thresholds tuned to match
• Decoding tests with real MP3, FLAC, Ogg and M4A files
• Firefox and Safari coverage
• Export of processed audio and direct PDF reports
• Optional source separation using a real model

Licence

Released under the MIT Licence.