Cleaning up a voice recording

I have a recording of a meeting at a restaurant, made through an android phone. Unfortunately, I can hear the background music, children babbling, cutlery clanging, mixers churning and scratching sounds MUCH louder than the voice recordings. Please help me clean it up! I cannot put even a sample of the recordings since they are confidential (office property), but I would appreciate any help I can get in cleaning this up since I have about 8 hours in total to transcribe!

I have tried the “Noise Reduction” feature, but they hardly help. The scratching sounds reduce to an extent, but the voices become tinny and un-decipherable

The recording is in WAV format, I am using Audacity 2.1.2 (ZIP download) on Windows 8.1 Enterprise.

but they hardly help.

It wouldn’t help. Noise Reduction is designed to gently reduce constant background noises such as air conditioners, fans or mild microphone hum. It won’t do a thing for your list of noisemakers.

I’ve used the VoiceMemo feature of my iPod with the unit sitting on the desk between all the participants and while the recording will not win any Hollywood production awards, it’s remarkable in its clarity.

Where was the android phone? If it was under the table or on the seat, you have the perfect conditions for a terrible restaurant conversation recording.


Audacity’s limiter can reduce momentary loud sounds so they are same volume as the quieter speech …
Demo of Audacity's Limiter.gif
That makes it more listenable, but not necessarily more comprehensible.

The bad news is, the human brain is often the best “filter”. The good news is, apparently you can understand what’s being said even though it may take more than one listen.

Nothing you do will make the recording comfortable to listen to, but you may be able to improve the intelligibility to some extent.

You can also try the Equalization effect to turn-down the low frequencies (maybe below 200Hz) which don’t contribute to intelligibility.

Turning-up the higher frequencies (maybe above 4kHz) will help to bring-out the “T” & “S” sounds, although that will bring-up the higher frequency noises too so you’ll just have to give it a try. And it might help to bring down the highest frequencies (maybe above 10kHz).

The main “voice frequencies” are around 200-300Hz so you can experiment with boosting the low-mid frequencies. But again, any noise in the same frequency band will be boosted too.

For this kind of experimenting, I find the the Equalizer’s Graph EQ mode easier than the Draw Curves mode.

I have about 8 hours in total to transcribe!

Wow! That could take 3 or 4 days!

I might help to split the recording into one-hour or half-hour segments (just to help with navigation).

Do you have any transcription software? There are applications that make it easier to fast-forward & rewind and to change playback speed without changing pitch. But if you don’t already have the software, this may not be the time to learn something new… You probably need to spend your time typing.