Loudness normalization for files with very low sound about ffmpeg-normalize HOT 3 CLOSED

Hard to reproduce without a sample, but I think R128 is not very reliable with short and low volume. Can you try setting simple peak normalization and see if that produces better results? (-nt peak)

from ffmpeg-normalize.

Alright so it actually seems like peak normalization does the trick, thank you! Though I'm no audio expert so I have no idea why. I was under the impression that loudness normalization was the way to go if the objective is to achieve a consistent volume for all files...

Here's a sample file:
https://github.com/slhck/ffmpeg-normalize/assets/32488097/2bb82dad-6f61-479f-a22b-e708d0e0f3d7

from ffmpeg-normalize.

Great to hear that peak normalization worked for you. The reason why EBU-type loudness normalization might not be that effective is that it is based on standard audio dynamics, but in your case, where the volume is extremely low, you'd have to apply some pre-gain for the algorithm to detect something useful.

Peak normalization simply amplifies the audio signal so that the highest peak reaches a certain level (e.g. 0 dBFS; can't get higher than that, otherwise it will clip). This can make quiet sounds more audible, but it does not consider the perceived loudness or dynamic range. It's a more straightforward method and can be more effective for files with very low sound like yours — when the loudnorm filter simply doesn't have enough gain to do anything useful.

If consistency across all files is your goal, you might have to use a combination of both methods. For example, you could use peak normalization to bring the quieter files up to a more standard level and then apply loudness normalization to the entire collection for uniformity.

Closing this for now, but if you encounter any more issues or have questions, feel free to reach out!

from ffmpeg-normalize.