A click and a ringing tone demand different views, even when both live in the same WAV file. A short analysis window makes timing easier to see; a longer window separates nearby frequencies more clearly. Colour range, channel choice and zoom can change the picture just as dramatically. Set up the display around the sound you heard, then keep those settings steady while comparing a repair.
Match settings to the audible symptom
Listen before converting a WAV to spectrogram form. Mark a click, whistle, rough vocal edge or strange tail with its time position and a few plain words. Include audio before and after that point. A view of the entire song can help locate sections, but it often hides the small event you need to investigate. Zoom to the phrase after you have a reliable listening reference.
Check the file’s sample rate and channels. The highest representable frequency is limited by half the sample rate, though useful signal bandwidth can be lower. A display set above that limit will not reveal extra sound. A stereo file also needs a channel decision: inspect left and right separately when the complaint is directional, rather than averaging away a difference between them.
The main controls are analysis window length or transform size, window function, overlap, frequency scale and magnitude range. Applications use different names and defaults. Record the actual values, not simply a preset called detailed. A preset may change several controls at once, which makes it harder to tell why a click or ring looks different.
A waveform remains useful alongside the spectrogram. It shows amplitude against time and can help locate an edit boundary or a sharp pulse. The spectrogram then shows how energy is distributed across frequency around that point. Neither view replaces listening. A visually obvious vertical stripe might be a legitimate drum hit, while a troublesome texture can be hard to see in a dense mix.
Use short windows for timing
Choose a relatively short window when you need to locate a click or determine whether an attack spreads before the audible hit. As a numerical example, 512 samples cover about 10.7 milliseconds at 48 kHz; 2,048 samples cover about 42.7 milliseconds. These are window durations, not promised display precision or recommended settings for every editor. The window function and display interpolation affect the appearance too.
A shorter window gives poorer separation between nearby frequencies. A narrow tone may look wider, and bass harmonics can merge. That tradeoff is acceptable when the question is whether a short unwanted event occurs between two words. It becomes less useful when you need to distinguish two close ringing frequencies. Switch settings to answer the second question rather than expecting one view to show everything.
Increase overlap if the tool allows it and the display appears too coarse in time. Overlap samples the analysis more often; it does not undo the time-frequency tradeoff of the window itself. A dense set of frames can make the picture smoother without providing the frequency detail of a longer window. Keep the audio selection unchanged while testing this control.
For a click repair, inspect the same area before and after processing with identical settings. Then listen for a new gap, dull attack or softened consonant. A smaller vertical mark may mean the click is reduced, but it can also mean the legitimate event was weakened. Keep enough surrounding sound in the audition to hear the edit as part of a phrase.
Use longer windows for persistent tones
A sustained whistle or metallic ring often benefits from a longer analysis window. More samples allow finer spacing between frequency bins. At 48 kHz, a 4,096-point transform has bin spacing of roughly 11.7 Hz. Bin spacing is not the same as guaranteed ability to resolve two real tones: the window shape and their relative levels still matter. Treat the number as a guide to scale.
The cost is less precise timing. A short tone can appear spread across a longer interval, and an abrupt onset may look gradual. A display may place each frame at its centre or use another alignment convention. Do not measure the exact start of a click from a long-window picture. Return to a shorter view and the waveform when the timing affects an edit.
Follow the ring through several notes or hits. A stable narrow band may support a restrained frequency-specific adjustment. A band that bends or follows the melody may belong to a musical source or a moving defect; a fixed notch will affect it unevenly. Compare a neighbouring passage where the sound is absent before selecting a repair frequency.
If both an attack and a long tail are troublesome, save two views of the same region. One can show the onset clearly and the other can separate the tail’s tones. Label their settings so the pictures do not appear to contradict one another later. They describe the same audio with different compromises, not two different versions of the sound.
Keep colour range consistent across tests
Colour usually represents magnitude within a chosen decibel range. A narrow range can emphasise weak residue; a wide range may make the same residue almost disappear. Neither setting changes the file. Keep the lower and upper limits fixed when comparing the original source and processed audio. Otherwise a seemingly cleaner image may come from the display rather than the repair.
Automatic scaling is convenient for exploration but troublesome for before-and-after evidence. If the processed file is quieter, the software may stretch its remaining energy back across the full palette. The result can look just as bright even after a real reduction. Conversely, different scaling may make an unchanged texture look dramatically better. Disable automatic adjustments when possible, or record exactly how they were applied.
Choose a frequency scale suited to the question. A linear scale gives equal display space to equal frequency intervals. A logarithmic scale spreads lower musical regions and compresses the top end, which can make harmonic relationships easier to follow. Use the same scale in both pictures. A squeezed upper band should not be interpreted as a missing high-frequency signal.
Level-match the listening comparison separately from the visual settings. If you apply gain to match the files, document whether the spectrograms show the original levels or those adjusted levels. Both approaches can answer useful questions, but mixing them obscures what changed. Keep the comparison simple: the same passage, channels, settings and clearly stated gain treatment.
Save settings alongside the comparison
For a wav spectrogram online tool, check which settings it actually exposes before relying on the image. Some displays offer a fixed transform size or automatically normalise each upload. They may still help locate an event, but a fixed or hidden setting limits repeatable comparison. Avoid uploading private or unreleased material unless the tool’s handling of files is acceptable for your work.
A local editor can make it easier to keep files aligned and revisit the exact selection. Save the project or note the start and end times, transform size, sample rate, window function, overlap, frequency scale, channel and colour limits. Include the original and processed filenames. A screenshot without these details is a memory aid, rather than a reproducible measurement.
Recheck the accepted export, because the file you deliver may differ from the preview inside the editor. Confirm its duration, channels and format, then inspect the marked event using your saved settings. Listen at comfortable volume with the display hidden for part of the test. The repair should still be useful when the picture is no longer steering your attention.
Keep the original file and stop changing settings once you have answered the practical question. A spectrogram can always be made more dramatic by zooming, recolouring or changing the window. The valuable result is narrower: a click located accurately, a ring understood well enough to test a small repair, and musical detail checked after that repair. Save that evidence with the working master.
