Spectrogram Online
See which frequencies are in a sound and when. Open an audio file or use your microphone, and the spectrogram shows time across, frequency up the side and loudness as colour. Choose the FFT size, a linear or log scale and the colours, point at it to read exact values, and save it as a PNG.
- 1Open audio or use the mic
- 2Adjust the view
- 3Save a PNG
1Open audio or use the microphone
or drop a file here · MP3, WAV, M4A, OGG, FLAC, or a video’s sound
Use microphone asks your browser for permission. The sound stays on this device.
Your audio never leaves your device. It is analysed in your browser. We count visits and finished downloads (tool name only) with analytics that sets no cookies.
2Adjust the view
How to see a spectrogram of audio
- Open your audio. Press Choose audio or drop a file on the box. Video files work too: the tool reads their sound. To see sound live, press Use microphone and allow it.
- Read it. Time runs left to right and frequency bottom to top. Bright colours are loud, dark ones quiet. Point at any spot to read its time, frequency and level, and the loudest frequency at that moment. Click to play from there.
- Adjust it. Switch between a log scale (more room for low notes) and a linear one, change the FFT size, the frequency and dB range, the time range and the colours.
- Save it. Download the spectrogram as a PNG up to 3840 × 1080 px, with or without axes, or the waveform of the same stretch.
What you get
- The file’s own sample rate. The tool reads the rate from the file and decodes at that rate, so the frequency axis goes up to the file’s real limit (half its sample rate): 22,050 Hz for a 44.1 kHz file, 24,000 Hz for 48 kHz.
- Levels in dB of full scale. A full-scale sine wave reads about 0 dB, and every halving of its level is about 6 dB less. A tone that falls between two FFT steps reads a little low: 0.6 dB low in our test below.
- Checked placement. We tested the drawing with made-up sounds whose frequencies we knew: steady tones at 250 Hz, 1,000 Hz and 6,000 Hz, a sweep from 100 Hz to 15,000 Hz, a 2,000 Hz MP3, and a 1,000 Hz tone fed in as a microphone. We saved 1200 × 400 px PNGs (FFT 2048) in Chrome, Firefox and Safari’s engine and found the brightest pixel in every column. On the linear scale it was always within one pixel row of the true frequency. On the log scale the typical error was under 10 cents (a tenth of a semitone); the largest, about one semitone, was on the lowest notes, where one FFT step (21.5 Hz at 44.1 kHz) is wider than a semitone, so use a bigger FFT for bass. The loudest-frequency read-out showed 250, 1,000 and 6,000 Hz, and a tone at half of full scale (true level −6.0 dB) read −6.0 to −6.6 dB.
- Live view. With the microphone on, the newest sound scrolls in on the right. Press Stop and analyse to keep what you recorded and look at all of it, then save it like a file.
Choosing the FFT size
Each column of the spectrogram is one FFT of a short slice of sound. A bigger FFT uses a longer slice, so it splits the frequencies into finer steps but smears fast changes. At 44,100 Hz, an FFT of 2048 gives steps of about 21.5 Hz and slices of about 46 ms. For speech and drums try 512 or 1024; for notes and steady tones try 4096 or more. The step size is shown under the picture after each change.
FAQ
What is a spectrogram?
A picture of sound. It shows time from left to right, frequency (pitch) from bottom to top, and how loud each frequency is as colour. A whistle makes a thin line, a voice makes stacked bands, and a drum hit makes a short vertical streak.
How do I see the spectrogram of my voice?
Press Use microphone, allow access, and speak or sing. The picture scrolls as you go. Press Stop and analyse to keep the recording, look at it all, play it back and save the picture.
Should I use a linear or a log frequency scale?
Log spaces octaves evenly, so 100 to 200 Hz takes as much room as 1,000 to 2,000 Hz. That suits music and voices. Linear spaces every hertz evenly, which shows high frequencies and harmonics in more detail.
Can I find hidden images in audio?
Yes, if someone drew one into the sound. Pictures hidden in songs or puzzles show up as shapes in the spectrogram. Try the linear scale and an FFT of 4096 or more, and set the floor to about −90 dB to bring out faint parts.
Which audio formats work?
Anything your browser can decode: MP3, WAV, M4A (AAC), OGG, FLAC and the sound of most MP4, MOV and WebM videos. Safari cannot open some OGG files. If a file will not open, convert it to WAV or MP3 first.
Is there a length limit?
No fixed limit, but the whole file is held in memory. On very long files the time step between columns is widened so the analysis still fits; the status line says when that happens. Microphone recordings stop on their own after 10 minutes.
Does this tool upload my audio?
No. Your file or microphone sound is analysed by your browser on your own device and is never sent to a server. You can check: open your browser's developer tools, go to the Network tab, and open a file. The page counts visits and finished downloads (tool name only) with analytics that sets no cookies, and no audio is ever sent (see privacy).