aa4a (Audio Analyzer for Android) is a real-time audio spectrum analyzer and scrolling waterfall (spectrogram) for Android. Tune an instrument, look at a voice, find the hum in a room, or watch what a microphone actually hears.
aa4a is a fork of woheller69/audio-analyzer-for-android. It keeps the analysis engine and rebuilds the display: the waterfall scrolls smoothly at the screen's refresh rate, the most-used analysis options are on the main screen, and the waterfall can fill the whole screen.
The published app, its F-Droid package, releases and translations belong to woheller69's repository. Go there for installable builds, to support his work, and to read his note on Google's 2026/2027 developer-verification requirement.
- Three views: spectrum only, spectrum over waterfall, or waterfall only. In the combined view, drag the divider to share the height between them.
- Smooth waterfall: a new row every ~21 ms by default and 30 s of history, drawn on every screen refresh with sub-pixel scrolling.
- Analysis options on the main screen, each showing what it costs you: FFT size (frequency resolution and window length), overlap (time per row), waterfall color range (with Auto), frequency scale, sample rate, window function, averaging, history length, palette, dB/dBA weighting, and cursor.
- Frequency axis: linear, logarithmic, or labelled by musical note.
- Cursor: long-press a plot to place a cursor across both plots, drag the cursor line to move it, and long-press again to remove it. The Cursor button sets its frequency exactly.
- Gestures: pinch to zoom, drag to pan, double-tap to reset. Frequency zoom is shared by the spectrum and the waterfall.
- Peak readout: peak frequency with interpolation, as Hz and as a note.
- A-weighting (dBA), although not suitable for serious measurement.
- Averaging of several spectra for a smoother plot.
- Record to WAV (PCM) while analyzing.
- Every recorder source that doesn't need root (MediaRecorder.AudioSource), and every sample rate the phone supports.
- Microphone calibration files, see the example.
- Test sources: two sine signals, white noise, and a repeating 50 Hz – 15 kHz sweep, selectable under Preferences → Audio source.
Requires the Android SDK (platform 35) and JDK 17. The Gradle wrapper (8.6) does not run on newer JDKs.
echo "sdk.dir=$HOME/Android/Sdk" > local.properties
JAVA_HOME=/usr/lib/jvm/java-17-openjdk-amd64 ./gradlew assembleDebug
adb install audioSpectrumAnalyzer/build/outputs/apk/debug/audioSpectrumAnalyzer-debug.apk
The app ID is org.xeyes.aa4a, so it installs alongside the upstream app
rather than replacing it. Debug builds add a .dev suffix and show up as
"aa4a dev" with a DEV badge on the icon, so a debug build and a release build
can be installed at the same time.
- Microphone, of course.
- External storage, only if you want to record to WAV.
This app does not send any personal or non-personal information in any form over the network.
Only with the user's permission and explicit request can this app store microphone data on the user's device.
The microphone permission is required because microphone data is what the app analyzes to compute and display the spectrum and waterfall.
The storage permission is optional and used only to save microphone data as WAV (PCM) files, for processing in another app. Removing recordings that are no longer needed is up to the user.
AnalyzerActivity: the controller. Lifecycle, permissions, touch gestures, and one setter per analysis option, each of which stores the preference and restarts sampling if it has to.ControlBar: the row of option buttons and their pickers.AnalyzerViews: the text readouts (peak, cursor, RMS, resolution, recording time) and notifications.AnalyzerGraphic: the plot view. It lays outSpectrumPlotandWaterfallPlotfor the current view mode and keeps their frequency axes in step.WaterfallPlot: history as a ring buffer of dB rows, colorized into a ring of 64-row bitmap tiles so a new row re-uploads one small tile. A display clock runs at the nominal row rate and locks onto actual row arrival, which makes scrolling smooth between rows. Zooming stretches the tiles, and they are re-rendered once the view has been still for a moment.SamplingLoopandSTFT: the model. Sampling, windowed FFT, averaging, peak and RMS.
The processing loop is run() in SamplingLoop.java. Each pass of
while (isRunning) reads a chunk of samples:
record.read(audioSamples, 0, readChunkSize);
and streams it into STFT.java:
stft.feedData(audioSamples, numOfReadShort);
which computes RMS and an FFT whenever enough samples have arrived. Each new spectrum is handed to the views with
activity.analyzerViews.update(spectrumDBcopy);
which stores it for the spectrum plot, appends it to the waterfall history, and requests a redraw. While the waterfall is visible it also redraws itself on every frame, so it keeps scrolling between spectra.
aa4a is the latest branch of a long line of work, and most of what it does was inherited.
- Stephen Uhler (thinkingcow) at Google, 2011–2012, wrote the original Audio Spectrum Analyzer for Android as an experiment with the Android SDK and published it on Google Code (see README.old). His sine generator, selector widget and FFT wrapper are still here.
- Eddy Xiao (bewantbe), 2014–2017, turned it into a real analyzer: the STFT engine, the spectrogram, gesture zooming, window functions, colormaps, calibration and most of the architecture.
- woheller69 (Wolfgang Heller) has maintained it since, published it on F-Droid, kept it building on modern Android, and extended it. This fork starts from his version 3.2.
- Contributors along the way: nfsmaster208, Steven Schoen, Hans-Joachim Zimmer, kennyzzhang, Yurt Page, Izzy (fastlane metadata), james34602 (additional window functions), and the translators on Toolate.
- FFT: FFTPACK by Paul N. Swarztrauber (NCAR, public domain), ported to C by Pekka Janhunen, to Java (jfftpack) by Baoshe Zhang at the University of Lethbridge, and adapted for Android by Stephen Uhler.
- Kaiser window Bessel function: from CERN's Colt library.
- Colormaps: viridis, magma, inferno and plasma from matplotlib, by Stéfan van der Walt and Nathaniel Smith (CC0). Parula is modeled on MATLAB's default colormap. The black-body map was made perceptually uniform with bewantbe's colormap_uniformize.
The scrolling waterfall, the Spectrum / Both / Waterfall views and the on-screen analysis controls were written by laplaces-agent, an LLM agent. Spectroid set the bar for how a waterfall should feel. It was watched and timed, not read.
An LLM can write this code only because people wrote code first. Before laplaces-agent wrote a line of this fork, the model behind it was trained on an enormous body of human work: source code, documentation, tutorials, answers to strangers' questions, bug reports, code reviews and mailing-list arguments. That work was written over decades by more people than can ever be counted, and few of them were asked. Many shared it freely and would be glad to see it go further. Some would not have agreed to this use if anyone had asked. Most were never given the chance to say.
Their names aren't here, and they can't be, because training doesn't record who taught the model what. So they are thanked here as a whole: every developer who wrote something down so the next person wouldn't have to work it out again. The ring buffer, the frame-synced render loop and the popup list with its checkmark didn't originate with laplaces-agent. They came from you.
Nothing in this fork was knowingly copied from a particular source. That doesn't settle the debt; it only describes it. The least that code written this way can do is go back where it came from, so this fork is released under the same free license as the work it builds on, for anyone to read, change and share. If some of your work is in it, it is still yours as well.
Released under the Apache License, Version 2.0. The full text is in LICENSE.
Copyright thinkingcow (Stephen Uhler), bewantbe, woheller69, laplaces-agent.