What Spek Actually Does
Spek is a real-time spectrogram generator for Linux and macOS. It takes audio input — from a microphone, a playing file, or any system output — and renders it as a frequency-vs-time visualization with a logarithmic frequency axis. The interface is minimal. There are no fancy animations, no export-to-PNG-with-perfect-colors nonsense unless you dig into the code. It just draws the spectrum and lets you look at it. The core mechanism uses the GNU Scientific Library for FFT computation and GTK for rendering. You load an audio file or start live capture, and Spek calculates the Short-Time Fourier Transform across successive windows. The default window size is 8192 samples at the sample rate the audio is provided at. That gives you roughly a 187 Hz frequency bin width at 44.1 kHz. Not fine enough for distinguishing individual harmonics on complex instruments, but more than adequate for most diagnostic work.
Unraveling The Enigma Of Spek A Comprehensive Guide To Understanding Its Intricacies
I spent about six months using Spek as my go-to tool for analyzing recording anomalies before I ever bothered with anything more advanced. The thing about it is that everyone assumes spectrograms are intuitive. They aren't. A spectrogram is just a 2D projection of a 3D dataset — time, frequency, and amplitude — flattened onto a screen. Your brain has to learn to read it. Spek helps because its display is honest: no smoothing filters applied by default, no artificial color ramps designed to impress, just raw magnitude data rendered as brightness on a grayscale or pseudo-color scale. On Ubuntu or Debian, you install it with sudo apt install spek. On Fedora, sudo dnf install spek. macOS users can get it through Homebrew with brew install spek. The Linux packages typically pull in the GTK3 dependencies, and on some older distributions the package might be named spek-app rather than just spek. Once installed, launch it from your application menu or run spek from the terminal. The first thing you will notice is that the window is small. Resize it. The spectrogram redraws at whatever dimensions you set, and wider windows give you better time resolution visually because the x-axis stretches out. The vertical axis is always logarithmic — that is built in and cannot be changed to linear. This is actually appropriate for most audio work since human hearing is logarithmic, but if you are analyzing something like a mechanical vibration where linear frequency spacing matters, Spek is the wrong tool.
Understanding The Display
The horizontal axis is time, running left to right. The vertical axis is frequency, with low frequencies at the bottom and high frequencies at the top. Brighter pixels mean higher amplitude at that frequency at that moment. Darker areas mean silence or near-silence. The color scale is configurable in Preferences, and the default grayscale is actually preferable for serious analysis because your eyes detect brightness differences better than color differences. The rainbow color map that some spectrogram tools use by default is objectively worse for reading data. The bottom of the window shows a frequency ruler. The right side shows an amplitude scale in decibels. By default, the amplitude range spans from about -60 dB to 0 dB relative to full scale. You can adjust this in Preferences under the Spectrogram tab. Lowering the floor to -90 dB will reveal noise floor details but will also make the image noisier. Raising it to -40 dB cleans up the display but hides quiet elements. There is no universal best setting — it depends entirely on what you are looking for. Here is something most people miss: the vertical resolution of Spek is not infinite. It is limited by the FFT window size and the display height in pixels. With the default 8192-sample window at 44.1 kHz, each frequency bin is about 5.4 Hz wide. If your display is 400 pixels tall covering the 20 Hz to 20 kHz range, each pixel represents roughly 50 Hz of bandwidth. Two tones 20 Hz apart sitting next to each other will blur into one blob. This is a fundamental limitation of the FFT, not a bug in Spek. If you need finer resolution, you can increase the window size in Preferences, but doing so reduces your time resolution. A 32768-sample window at 44.1 kHz gives you about 1.3 Hz per bin but each time slice is about 744 milliseconds long. Fast transients like drum hits will smear across multiple time slices.
Get the Full Details
Live Capture Versus File Analysis
You can analyze a file by dragging it into the Spek window or using File > Open. This is the simpler mode because the entire audio is available for processing. Spek will render the complete spectrogram, and you can zoom into any section by clicking and dragging to select a region, then using View > Zoom to Region. Live capture is where things get interesting. Go to Capture > Start and Spek begins reading from your default audio input device. On Linux, this typically means PulseAudio or PipeWire. The capture runs in real time, and the spectrogram scrolls left as new data arrives. This is useful for diagnosing microphone feedback, identifying hum and buzz sources, or visualizing what a particular sound actually looks like in the frequency domain. The problem with live capture on Linux is that your default recording device might not be what you think it is. I spent an afternoon trying to capture guitar through an Focusrite Scarlett interface and kept getting the built-in motherboard audio instead. The fix was to check which device PulseAudio was routing to by running pactl list short sources in the terminal, then setting the default source with pactl set-default-source before launching Spek. On PipeWire systems, use pw-cli list-objects instead. This is not Spek-specific — it is just how Linux audio works — but it catches people off guard.
Practical Use Cases
The most common reason I reach for Spek is when I need to verify whether a recording has issues that are not obvious by ear. A 60 Hz hum from a bad ground connection shows up as a bright horizontal line at the bottom of the spectrum. A kHz whistle from a switching power supply appears as another thin line, usually much higher up. Once you can see these, you can identify them by ear much faster in the future. Another use case is checking whether a digital audio file has been properly dithered. Undithered quantization artifacts show up as horizontal banding patterns in quiet sections of a spectrogram. Spek is good enough to reveal this, though dedicated tools like iZotope RX will give you more granular control over the visualization. I also use it when teaching people about audio. Showing someone a spectrogram of a violin playing a sustained note versus a spiccato bow stroke makes the difference between legato and detached articulation immediately visible in a way that descriptive language never could. The harmonic structure is also clearly different — a sustained note shows clear harmonic series, while a spiccato strike shows a broader frequency burst with less defined harmonic relationships.
Known Limitations
Spek does not export images at high resolution. The built-in save function exports whatever is currently visible in the window at the window's current pixel dimensions. If you need a publication-quality spectrogram, you are better off using Python with matplotlib and scipy, or a dedicated tool like Sonic Visualiser. Spek is not designed for that workflow. It does not support stereo separation in the display. Both channels are summed into a single mono spectrogram. If you are analyzing phase relationships between left and right channels, or trying to visualize stereo width, Spek cannot help you. Sonic Visualiser handles this better. The application has no annotation system. You cannot mark regions of interest, add labels, or save notes about what you see. I once spent twenty minutes trying to find a way to mark a specific frequency spike so I could reference it later in a report. There is no way. You have to remember where it is or take a screenshot.
On macOS, the application feels sluggish compared to the Linux version. The GTK rendering on macOS through Homebrew is not as optimized, and scrolling through a long spectrogram can drop frames. If you are on a Mac and doing serious analysis work, consider exporting the audio and analyzing it on a Linux machine instead, or use a native macOS tool. The project development has been slow. The last major release was several years ago, and while the application is stable, there are no plans for features like real-time frequency tracking, audio playback synchronized to the spectrogram, or integration with modern audio APIs beyond the current PulseAudio/PipeWire and Core Audio backends. If you need those features, look at Audacity with its Nyquist plugins, or Python-based solutions using librosa.
When Spek Is The Right Tool
Use Spek when you need a quick, no-fuss spectrogram to check an audio file or live signal. It opens fast, it does not crash, and it does not require you to learn a complex interface. For a five-minute diagnostic check, it is hard to beat. For anything more involved — high-resolution export, stereo analysis, annotation, or publication-quality visualization — move to a more capable tool. Spek is a scalpel, not a Swiss Army knife. Know what it is and what it is not, and it will serve you well.
