Narration arrives from a subject matter expert, a freelancer and an internal team, and each one is fine on its own and jarring in sequence. This measures every file and reports the spread between the quietest and the loudest, which is the number that decides whether a learner reaches for the volume control between slides.
Free · No account · Your files never upload
Every file is measured on your computer. Nothing is uploaded and nothing is changed.
Measures the peak and the average level of every file, in decibels relative to full scale, and reports the spread between the quietest and the loudest.
No single file is the problem. A course is a sequence, and a learner notices the change between slides rather than the absolute level of any one of them. Anything past about three decibels of spread is audible as a jump, and it is what people are describing when they say a course sounds badly produced without being able to say why.
Peak is the loudest single moment. RMS is the average across the whole file, and it is the one that corresponds to how loud something actually sounds. Two files can share an identical peak and be obviously different in volume, which is exactly why normalising to peak alone does not fix this problem, and why plenty of tools that claim to normalise leave you no better off.
These are RMS figures, not LUFS. Proper loudness units need the weighting filters from the broadcast standard, and calling an RMS number LUFS because it looks similar would be wrong in a way that only shows up when somebody checks it against a real meter. For the question this answers, which is whether these files sit together, RMS answers it correctly.
Clipping is detected but cannot be repaired. Once a waveform has been squared off at full scale the information is gone, and those files need re-exporting from the original at a lower level.