Audio hiss revisited

Yesterday's podcast was a test of a new procedure for telephone recording on a shoestring. When I earlier complained about the problems I'd been having with audio hiss, people suggested all kinds of remedies. My audio guru Doug Kaye, of ITConversations, offered no less than three: a hybrid coupler (he recommends Gentner), this Skype recipe, and this two-phone-line hack. For now, though, I'm hoping that the technique I'm using will suffice, because it's simpler and/or cheaper than anything else I've found.

In my case, moving the analog-to-digital conversion out of my PC and into an external sound card does seem to have been a decisive factor. I bought a USB Audio UA580 for under $30. Its 5.1 Channel support wasn't necessary, but I'm glad it has both line-level and microphone-level inputs -- that means I can use it with my Mac as well as my PC. When I routed the QuickTap through the UA580, the signal became much cleaner.

There was still a little noise, which made me realize that the problem I'd been having was a combination of computer noise and telephone-line noise. Removing that combination was asking too much of Audacity's noise-removal filter. But by factoring out computer noise, it seems that I'm now able to use Audacity's noise-removal effectively, without creating annoying artifacts. Two aspects of that procedure are worth mentioning. First, I use the least aggressive noise-removal setting. Second, when sampling noise to remove, I gather samples from points throughout the recording to account for variation in telephone-line conditions.

Then there's the question of audio levels. Because a cheap coupler like the QuickTap hears a blended mix of the signals from the handset cord, there's no way to separate my voice from the interviewee's and independently adjust the two. A high-end hybrid, Doug Kaye says, "actually subtracts your voice from the phone line in order to isolate the two voices" -- and without one, "your voice will be much louder than your interviewee's." Over the course of a few interviews, though, what I observed was a lot of variability. On some calls my voice was louder than the interviewee's, but on other calls it was softer. There can also be quite a bit of variation in either voice during a call -- even, in my case, during a sentence, because I've found that I tend to start loud and trail off. So here's my low-tech way of addressing this problem. If the interviewee gets too loud or too soft, I'll say so, and then edit that instruction out later. Meanwhile I monitor my own levels as I'm speaking, and tweak them by adjusting the distance between phone and mouth, or the loudness of my voice, or both. It's awkward, to be sure, but perhaps no more so than twiddling knobs while talking. If you're the interviewer and the producer, you can either do this live or in post-production. And for the kinds of things I'm interested in capturing, life's too short for a lot of post-production.

Finally, I changed my procedure for encoding MP3 files. I'd been exporting them directly from Audacity, and I'd wondered why I kept having to bump up the bitrate beyond what I thought it needed to be -- to 96kbps or more -- to get reasonable quality. So I exported to uncompressed .WAV instead and tried using lame directly from the command line. It appears that when Audacity calls the lame library you get default quality, not the slower but better results you get with the -h command-line flag. There's probably a way to get Audacity to use that mode but, for now, I'm fine with saving uncompressed audio and encoding from the command line using -h, which appears to give me much better results at lower bitrates.

Ultimately, of course, a voice call -- analog or digital -- -- is a dicey medium for recording. That's why, Doug Kaye informs me, Moira Gunn's excellent Tech Nation show is done entirely in-person. If you want studio quality, do it in the studio. But the reality is that I can't usually sit face-to-face with Tim Howes or Phil Libin or Jeff Nielsen. I can, however, call these folks and capture some interesting conversations. No audiophile will think that yesterday's podcast sounds great. But if it's good enough, I'll be satisfied.

Former URL: