Turning a recording up can make it easier to hear, but it also turns up the noise and the loudest peaks. Once those peaks go past the output range, the saved signal can clip. A louder file isn't automatically a clearer one.
Volume Changer and Normalizer shows the waveform, checks the decoded peak and RMS level, and marks regions that would exceed full scale at the selected gain. Use those checks alongside your ears.
Try a signal with a known starting level
The sample tone is a generated 440 Hz sine wave, two seconds long. It's a mono, 44.1 kHz, 16-bit PCM WAV with a peak close to one quarter of full scale. It's a test signal, not a recording of someone speaking. Turn your volume to a comfortable level before playing it.
Open it in the volume tool. The page reports a peak of about -12.0 dBFS and an RMS level of about -15.1 dBFS. Those values are from Chromium 146 on Windows. The peak is the largest decoded sample. RMS is the overall level, based on the samples' squared amplitudes. They're different measurements.
In Manual gain, choose +6 dB. That roughly doubles the sample amplitude, which brings this tone's peak to about half scale. Choose +18 dB and the waveform warns that the result would clip. That's roughly eight times the amplitude, which puts a quarter-scale peak well past full scale.
The red marks are a prediction based on the decoded samples and gain. They don't fix anything, and they don't measure true peaks between samples in an encoded delivery file. Check the export too, especially when you're making a lossy listening copy.
What peak normalization does
Select Normalize peak on the same source. This mode finds the decoded peak and calculates one gain value that brings it to 0.99 of full scale, leaving a little headroom. For this tone, the displayed gain is about +12.0 dB. The waveform no longer shows the +18 dB clipping warning.
Choose WAV, select Apply and export, and download the result. It's named quarter-scale-tone-plus12p0dB.wav, with the gain rounded to one decimal place. The WAV export uses 16-bit PCM at 44.1 kHz with two channels. MP3 export uses 192 kbps.
The mono sample also shows why you should check the export separately. Mapping it into the WAV output's two channels changes the level of each channel. In the saved file, each channel peaked at about 0.70 of full scale, roughly -3.1 dBFS, instead of the 0.99 target. FFmpeg's resampler documentation covers channel layout remapping. A gain calculated from the decoded input doesn't tell you what the final file measures.
Peak normalization applies the same gain to the whole file. It doesn't turn up a quiet sentence separately from a shout. One loud click can set the peak and leave the rest of the recording quieter than you expected. So look at the recording before picking a mode instead of normalizing every file by default.
The presets aren't a LUFS meter
Podcast level, Streaming level, and Quiet level use approximate RMS targets of -20, -14, and -26 dBFS. The tool works out the difference from the decoded RMS and applies that gain, within its gain limits. It doesn't run an ITU-R BS.1770 loudness measurement or an automatic limiter.
So a track with quiet passages and sharp peaks can hit a preset's average target and still push its peaks into clipping. Watch the red regions after changing modes. If a delivery spec calls for integrated LUFS or true-peak limits, use software that measures those values. These presets won't tell you whether you meet it.
Check the useful part of the recording
For speech, compare a quiet sentence and a louder one at the same listening volume. Listen for background noise getting distracting, not just the voice getting louder. Noise reduction is a separate job, and more gain won't bring back detail that was clipped in the original.
Waveform Image Generator can give you a visual reference, and Tone Generator is handy for making other controlled signals. A waveform still doesn't replace listening to the saved output.
Keep the source and export a copy. The audio format guide explains why converting a clipped or compressed recording to WAV doesn't bring back what was already lost. Start with modest gain, check the peaks, and listen to the file you're going to share.