This is very likely in the tutorials but I just cannot find it, probably because as a total sound analphabet I dont know the names of things:
In a spoken word recording, to amplify a voice that is far away from the mic and almost inaudible, I use the compressor on that piece (default settings, played a bit but did not get far), then reduce the noise such that the level of the background noise matches the rest of the recording.
To get the noise profile, I need to deselect the original selection, but then I need to find the exact selection again. How do I bookmark/label the selected piece? I’m sure I can just mark a stretch of audio and select it again by just clicking into the mark.
If there is a better way to amplify quiet voices, advice is appreciated. Compressor distorts but works reasonably well for the large amount of amplification this sometimes needs, with the voice completely buried in the noise and audible but not visible.
Audacity 2.0.5 on macbook pro running 10.8.5
thanks!
to be more specific: There is one quite good AKG mic in a room the size of a smaller seminar room. Low voices are nicely audible for everyone there, just on the recording they can be very quiet if the person sits farther away and talks quietly. The recordings are generally good quality, even though they are only 96 kbps stereo mp3 (exiftool, I don’t do the recording). Otherwise I guess the compressor would not be able to get them out.
Might it be better to reduce the noise before compression?
COMMAND + B to label the selection - see Label Tracks .
Select a piece where you hear the quiet voice then look in Analyze > Plot Spectrum… . Compare with the Spectrum of the noise without the voice speaking.
If there are any spikes in the spectrum of “voice and noise” that are not in the spectrum of "noise only), you could try Effect > Equalization… ( Audacity Manual ) to increase the frequencies observed in the spike and reduce all other frequencies. But it may make little audible difference.
hm, if I do it that way round, it sounds like what might be a longer attack/decay time, like a lag in the response, each word fading in.
thanks! I dont know how this escaped me.
wow, this is nice. Much less muffled than the compressor. Spectrum has higher peaks between 4000 and 12000 kHz. Modified the Treble Boost curve to make one square bump to +12 dB from 5000 to 10000. Starting at 4000 would add something weird, and going higher than +12 dB lifts the noise up too much, but this just makes the voice clearer = understandable even if quiet, and I don’t even need noise reduction that often brings up the ess too much. I could even play with careful Equalizer over the whole recording. But I don’t want to lose the warm tone of the voice.
Amplifying a voice that is “far away” will never give optimal results - at best you are “making the best of a bad job”.
Generally I use noise reduction before compression, and I find that this usually gives better results than the other way round.
The problem with applying compression first is that it causes fluctuations in the noise level. During loud parts the noise floor will be lower. During quiet parts the noise level becomes higher. The effectiveness of noise reduction is highly dependent on how well the “noise profile” matches the noise that you wish to remove. If the level of the noise that you want to remove is going up and down then it is impossible to closely match all of it. (The same applies when using dynamic range expansion).
Occasionally you may find that compression first produces a (subjectively) slightly better result, so for important work it is worth trying it both ways, but in my experience the best order of effects (99% of the time) is;
Amplify (optional, but it’s easier to hear what is happening if the audio is not too quiet)
Noise reduction (if required)
Dynamic processing (compression/expansion) if required.
As a “rule” I would say “noise reduction before dynamic processing”.
that makes perfect sense. Maybe this noise floor fluctuation is what sounded to me like “fade-in”
subjectively, that too. I hear older edits of mine and think, what was I doing here, because I had not heard the distortion that I had introduced. Still learning.
Sounds good to me. Also the “If required”, often it would probably be best to just leave it noisy.
I remember hearing this Opus Kura edition of the Casals Bach Suite recordings from shellac, suddenly there is a real cello there, before hearing that you don’t even know what the widely sold CD editions cut away together with the crackling.
Update:
the recording crew got a new mixer, and I can now use the Compressor on the whole recording without the louder ess becoming sharp. yippie, this makes things much easier. I can almost make a File > Chain now.
The new recordings are much better quality, I find. The dynamic space (is that the word? like the 256 grey levels in images) is nicely filled and nothing is clipped. I hate clipping, it hurts with earphones.
The normal words are a bit too low now, around -18 dB, so I compress the whole thing once or even twice 1:10 to bring the normal peaks up to -12 dB, and it still sounds natural to me. Maybe 1:10 is extreme but it seems to work here. It brings up the quiet voices nicely too.
Before compression, a noise reduction of -3 dB is fine, afterwards normalize and that’s it. Another key for quiet voices was mp3 export with variable bit rate: I can afford to leave it stereo, and use mp3 export as VBR Joint Stereo. quality 8 works nicely.
If something sounds wrong here, corrections are appreciated. I need to read some more on the compressor, I’m slowly getting it but it is sometimes difficult to hear differences.