Easy ways to improve sound quality on voice recordings?

April 16, 2013 7:06 AM

I have some digital audio recordings of interviews conducted by telephone that I must transcribe. The sound quality of the responses is very poor. In the past, I have been able to get through the poor quality and make out what was being said. However, these are particularly bad. I know very little about audio editing. I have tried using tools like Audacity to manage the sound levels and remove background noise, with some success. However, I wonder if there are other quick fixes that I could try?
posted by Grinder to Technology (7 answers total)

I'm doing telephone recordings, and I can tell you that if it's the quality of your phone line, there is no quick fix. I also use Audacity, and I tried equalizing it out, but that's tricky because you can also reduce quality by killing harmonics that give the sound richness. I had a hum that was caused by an induced +15 volts by nearby power lines and the phone company had to install a lunchmeat sized filter outside to eliminate it. Also, I tried to interview somebody who was using Google Voice and I ended up asking them to call me on their cell phone. Even though cells are notoriously bad for dropping, the quality was much better. While I was listening to it snap and pop, I was thinking, "Can I clean this up later?" The clear answer was no. Telephone connections aren't the best, but I think two old recording axioms apply: "Garbage in, garbage out", and you can't "fix it in post". A phone duplex box called an AutoHybrid might help you make it a little better to manage for future recordings.
posted by CollectiveMind at 7:48 AM on April 16, 2013

Audacity really isn't bad for this sort of thing, IMO, and given that I don't know your familiarity with audio concepts or computer recording in general, moving to a different program might present you with a fairly steep learning curve for using a specific program or for some basic understanding of audio concepts - in other words, you might get good or better results with a different program, but it might not be worth the time & effort to learn how to get those results.

Standard U.S. phone lines really only have a frequency range of 300 Hz to 3 kilohertz, low to high. Try using the EQ (equalization) to eliminate sound below 300 and above 3k to help eliminate unwanted sounds in the recordings.

The human voice tends to have a sort of "clarity peak" from about 2k to 4k, maybe try a little boost in that range.

If it sounds "boxy", try reducing somewhere in the 500 to 1k range, if "muffled" or "thick" try reducing 250 to 400 Hz.

And if you're not doing this already, try listening with headphones or earbuds to prevent sound from your physical environment interfering with what you're listening for in the recordings.
posted by soundguy99 at 8:19 AM on April 16, 2013

I found out about Auphonic on a recent FLOSS weekly podcast . I have now used it for digital recordings in a room where the questioners were sitting all over the place so there were many different sound levels.
I do use audacity for "Noise Removal" but i do not know if i need to even do that.
As a first pass to get something reasonable for editing or passing on to other people i upload the file and leave all the settings at default. I have it set up to download to my dropbox folder so it really is upload and leave. free account as well.
posted by stuartmm at 8:58 AM on April 16, 2013

I uploaded a snapshot image here to show you what the default settings will do to a file. It may not line up exactly as i cut off a few seconds at the start and a longer part at the end before i processed the top file. The auphonic output is the bottom 2 channels.
posted by stuartmm at 9:29 AM on April 16, 2013

Sorry if this is obvious, but are you listening with some quality headphones or speakers?
posted by unmake at 11:15 AM on April 16, 2013

You could cut out everything below 60hz...that tends to work pretty well for my field recordings, anyway.

Also try the audiotuts tutorials for equalization techniques!
posted by lhude sing cuccu at 2:57 PM on April 16, 2013

Soundguy99 offers excellent advice.

Keeping in mind that the goal is not to produce something which sounds good, but rather something which is maximally intelligible, you might also try experimenting even further with filtering and and or pitch adjustment if that isn't enough. (In audacity, you can draw arbitrary filter curves using the "equalization" effect, and you can try pitch shifting with the "change pitch" effect. Don't forget to make backups copies of the raw data first.)

When trying to pull intelligible content from ham radio transmissions with a lot of broad-band noise, I've found that filtering extremely aggressively can be useful. Don't be afraid to try a bandwidth of a couple hundred hertz or less, centered at a few different places throughout the band. The voices will sound very unnatural and unpleasant, but you may have a better chance at understanding them. (Humans seem to be better at filling in the missing out-of-band data from a very narrowly filtered voice than we are at ignoring noise in the same spectral region.)

I've also sometimes found that filtering aggressively with a narrow band pass filter at the low end of a voice and then shifting the pitch to higher frequencies seems to help. Whether it's due to some complicated psychoacoustic voodoo or just a placebo, I can't say. . . but, it's worth a try.

Another option is to try is plotting the spectrum (Analyze->Plot Spectrum, in Audacity) for a chunk of audio where there is no speech, and for another chunk of audio where there is speech. That may give you some clues about what to remove with your filters, if it isn't obvious when listening.
posted by eotvos at 6:57 PM on April 16, 2013