Open an audio file or microphone below to inspect its current frequency content. The audio is decoded and measured in this browser; it isn't uploaded, and the page doesn't send filenames, samples or measured values to analytics.
Start with the logarithmic view and a 4096-point FFT. That's a readable general-purpose setup for speech, music, hum and resonances. Increase the FFT size when two nearby steady tones blur together; reduce it when you need the display to react faster to short events.
Free browser FFT tool · no audio upload
Audio spectrum analyzer
Open a local file, use the microphone, or run a 440 Hz test. Inspect frequency peaks in real time. The 440 Hz test is silent; selected files play through your output.
Analyzer settingsFFT size, scale, channel and smoothing
Peak tableSpectral peaks; note labels may identify harmonics
Strongest spectral peaks
| # | Frequency | Level | Nearest note |
|---|---|---|---|
| — | Waiting | — | — |
How to analyze an audio file or microphone
- Choose the source. Open a browser-decodable audio file, allow microphone access only when you need a live input, or run the internal 440 Hz / A4 test to check the display path. The internal source does not test file decoding, microphone calibration or external recording accuracy.
- Set the question. Keep the logarithmic scale for musical balance, hum and broad tonal problems. Switch to linear when equal spacing in hertz matters more than equal spacing by octave.
- Choose FFT size. Use 4096 as a practical start, 8192–32768 for narrower steady peaks, or 2048 for quicker motion. The displayed Hz-per-bin value shows the actual spacing.
- Read the shape before the number. A single narrow spike, a harmonic series and a broad high-frequency shelf point to different problems. Hover or touch the graph only after you identify the pattern.
- Freeze or export the evidence. Freeze holds the current curve, not playback or microphone capture. Peak hold keeps the highest observed value in each bin. PNG captures the graph; CSV includes all non-DC bins up to Nyquist, including frequencies outside the plotted range. Press Stop to end the source.
Read a spectrum without guessing
| Pattern | Likely meaning | Next useful check |
|---|---|---|
| One stable narrow peak | A tone, electrical whine or strong resonance | Check whether harmonics repeat at exact multiples |
| 50 or 60 Hz with multiples | Mains-related hum is plausible | Compare 100/120 Hz and higher harmonic lines, then inspect the recording chain |
| Wide energy across many high bins | Hiss, air or broadband noise | Compare a noise-only region with wanted programme |
| Regular harmonic ladder | A pitched source or nonlinear distortion | Read the fundamental, then compare harmonic levels rather than deleting every upper peak |
| Curve changes every frame | Normal for speech, music and transients | Use smoothing or freeze; do not treat one moving frame as a full-file average |
A spectrum shows level by frequency at the current moment. The strongest peak can be a harmonic rather than the fundamental, so its nearest-note label may be an octave or another interval away from the played note. It can't tell you whether the sound is musically right, and it won't show when a frequency occurred across the whole file. That second job belongs to a spectrogram, where time is another axis.
What FFT size changes
The browser's analyzer uses a sample window whose length is the selected FFT size. The bin spacing is the context sample rate divided by that size. At 48 kHz, a 4096-point FFT spaces bins by about 11.72 Hz; 16384 narrows that to about 2.93 Hz. The graph stops at the lower of 20 kHz or the context's Nyquist frequency, which is half its sample rate. The larger window separates nearby steady components more clearly, but covers more time and responds less tightly to short events.
This is a tradeoff, not a quality slider. The MDN documentation for AnalyserNode.fftSize defines the supported power-of-two range, while the Web Audio visualization guide explains why the number of reported frequency bins is half the FFT size.
Smoothing changes the display over time. A high value calms movement and makes a stable tonal balance easier to read, but it can hide a short click, so lower it when the event is brief. Smoothing doesn't repair or filter the source. Peak hold can retain a source-start transient, so let a steady tone settle and press Reset peaks before judging its spectrum.
Why the graph says dBFS, not sound-pressure dB
The displayed level is the magnitude of an FFT bin, not a time-domain sample peak, RMS level or true peak. The Web Audio analysis method applies a Blackman window, a normalized transform and smoothing before converting magnitude to decibels. Bin position, windowing and smoothing affect the number. Do not compare it directly with a full-file loudness or peak meter. A microphone adds its preamp, device gain and system processing to that uncertainty; without a calibrated chain this is not dB SPL.
Use the microphone mode to locate relative peaks, harmonics and changes. Do not use it to certify workplace noise, speaker sensitivity or room SPL. A phone and a measurement microphone can show similar frequency shapes while reporting very different absolute levels.
What “local” means here
The file chooser creates a temporary browser object URL; Web Audio reads it in the page. Microphone mode uses the permission-controlled media stream. The implementation has no audio upload endpoint, and analytics records only coarse actions such as starting a source or exporting a graph.
Format support follows the browser and operating system. This tool gives the selected file to an HTML audio element and starts analysis only after the browser reports that it can play the media. The MDN documentation for HTMLMediaElement.canPlayType() explains why support is reported as “probably,” “maybe” or unsupported rather than as one universal format list. PCM WAV is the safest troubleshooting input when a compressed file refuses to open.
The channel selector changes the analysis, not the file’s speaker playback. Mix / mono sums the available channels, so opposite-polarity stereo can cancel. Left and Right inspect the first two channels separately; Right can be silent on a mono source. Use Mix / mono for mono recordings. For a full-file loudness or true-peak check, use the LUFS meter; for another measurement job, choose from the audio tools hub.
Keep your system volume low before selecting a file: file mode plays the recording while analyzing it. Microphone mode is not sent to your speakers, and the internal 440 Hz test is silent. If the graph stays empty, choose Mix / mono, unfreeze it and check the source. For microphone access, use HTTPS and allow permission in the browser; a permission request can remain pending until you answer. Stop cancels this tool’s pending start. For a file that will not open, try a short PCM WAV; use Play if mobile autoplay is blocked.
Continue the measurement in Sound Forge
For a single voice or instrument’s pitch, use the monophonic pitch detector and check its confidence rather than reading the highest FFT peak as a tuner. For work inside the editor, the Sound Forge Spectrum Analysis guide covers its menus, snapshots, monitoring and Sonogram.
Once you identify a stable line or noisy band, move to the repair that matches the pattern. Use the hiss and hum workflow for steady electrical lines or broadband noise, and the Sound Forge equalizer guide when a measured resonance needs a restrained, listenable correction.
Frequently asked questions
Is my audio uploaded to SoundForgePro?
No. The selected file, microphone samples, filename and measured values remain in this browser. Tool events contain an action label, not audio content, filenames or readings. Ordinary site analytics may still use browser or session identifiers; local audio processing is not a promise of anonymous browsing.
What is the best FFT size for audio?
There is no universal best size. Start at 4096. Increase it for closely spaced steady frequencies; reduce it for faster response to short events. Use the displayed Hz-per-bin value to judge frequency spacing.
Why is the microphone reading not dB SPL?
The browser receives a level shaped by the microphone, preamp, device gain and operating-system processing. Without calibration, it can compare relative spectral energy but cannot produce a defensible sound-pressure level.
What is the difference between a spectrum and a spectrogram?
A spectrum plots level against frequency for the current moment. A spectrogram adds time, so you can see when each frequency appears. This tool is a spectrum analyzer, not a full-file spectrogram.
Why does a 440 Hz tone show nearby frequencies too?
An FFT divides a finite sample window into bins. A tone that does not land exactly at a bin center spreads energy into neighboring bins, and the analyzer's windowing shapes that leakage. Read the dominant peak and bin resolution rather than expecting one isolated pixel.
Can the analyzer remove hum or noise?
No. It identifies frequency patterns but does not process or download modified audio. Diagnose first, then make a restrained repair in an editor and compare the same passage again.
Last fact-checked: September 13, 2026 against the W3C Web Audio specification, MDN documentation and this analyzer’s implementation.