Dynamic amplification of a sound file

Question

I have a 60-minute WAV file of a conversation. Here is a 600kB clip from it:

https://www.dropbox.com/s/m00d525v34bwbpa/needs-levelling.wav?dl=0

One voice is very quiet and I would like to amplify it.

I first tried Audacity's Leveller and Limiter effects, but they didn't amplify enough.

Then I tried ffmpeg's dynaudnorm filter, using various options:

ffmpeg -i needs-levelling.wav -af dynaudnorm tmp.wav 
ffmpeg -i needs-levelling.wav -af dynaudnorm=p=0.5 tmp.wav 
ffmpeg -i needs-levelling.wav -af dynaudnorm=r=0.6 tmp.wav 
ffmpeg -i needs-levelling.wav -af dynaudnorm=m=50 tmp.wav 
ffmpeg -i needs-levelling.wav -af dynaudnorm=m=100 tmp.wav 
ffmpeg -i needs-levelling.wav -af dynaudnorm=r=.95 tmp.wav 
ffmpeg -i needs-levelling.wav -af dynaudnorm=p=.01 tmp.wav 
ffmpeg -i needs-levelling.wav -af dynaudnorm=p=.99 tmp.wav 

However, none of these seemed to work either.

I would like to use ffmpeg or Audacity instead of installing a new program. Can anyone suggest what I'm doing wrong? Thanks.

Follow-up: Using Adobe Audition CS5.5 "Speech Volume Leveller" with the "Strong" preset worked great.

Before: https://www.dropbox.com/s/m00d525v34bwbpa/needs-levelling.wav?dl=0

After: https://www.dropbox.com/s/plmtygdluual8rb/needs-levelling-fixed.wav?dl=0

Here's the Speech Volume Leveller:

Levels before:

Levels after (amplified the quiet part in the first half!):

In contrast, Audacity 2.1.2 has a "Leveller" effect with similar options, but doesn't correctly amplify the quiet parts:

Before:

After (not much change to the quiet part in the 1st half!):

Answer 1

There is no magic trick to make something good out of a bad recording.

It will always sound very bad. That is why is it extremely important to get a good recording at first, with plenty of dynamic and the full spectrum of the sources: you will save yourself the pain of desperately trying to get something good out of a bad recording (Something you can't). If it's bad, just do it again; it's that simple.

However, if you are into this particular situation and you just need to get an understandable speech, and according to your example, you could try a bunch of different things:

• By using some gain, a gate and a compressor, you can make the quiet voice louder. You will also get a lot of noise, it will sound awful, but it will be understandable. Also use an EQ to suppress all the frequency into which your voice is not present, it will lower the noise and the air conditioner.
• You can also check the plugins from Izotope (RX5), they have developed different algorithms in order to lower the noise level (they analyze and then add the opposite wave of the estimated noise). Again, no magic here. You will get some audio artifacts if you go too hard.
• Finally, there is a very technical, and probably highly difficult solution, which is to extract the dialog and the intonation, and then reconstruct an artificial speech. We are now able to create an artificial human voice that cannot be detected by people (using neural network). Perfect result, but heavy investment.

You will need some decent audio software like Reaper or Ardour. In the end it depends on the time/energy/money your are ready to throw at it.

Final word, if you want to have a good result, just get a good recording. Basic holy rule of the audio guy.

Answer 2

Best choice inmho would be to edit the quite parts individually adding gain and reducing noise (signal-to-noise ratio is not favorable in the example). Better than the two options would be to redo the recoding or part of it if possible.