[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Fwd: [AUDITORY] 60 Hz hum causing problems with speech recognition
On Tue, Jun 16, 2009 at 10:43, Tarun Pruthi <t.pruthi@xxxxxxxx> wrote:
> Hi all:
> I have been having a lot of problems lately with 60 Hz humming noise in my speech recordings (I will confess that most of my tests have been on windows/linux laptops - and my guess is that they probably have much worse recording quality than desktops on average - but we live in a mobile world now and I have to make it work on laptops). Whenever it is there, it causes a serious drop in speech recognition accuracy, especially since it comes in very randomly at times giving for example clean recordings during speaker enrollment and noisy recordings in live testing. The problem is even worse on linux which has given me all kinds of artifacts in the recordings besides the 60 Hz hum - my guess is that this is most likely because of driver problems.
Hi, If you have issues on linux that you don't have under windows, I
would most likely examine your 'chain' of audio. Linux allows you
(generally) to access the sound hardware pretty directly, if you know
where to look. This is all dependent on your distro, your soundcard
and your individual setup, so I can't specifically help on that.
Generally, with all desktop AND laptop computers, I would recommend
getting an external USB soundcard. Get the A/D's out and away from
the noisy motherboard. USB drivers are fairly standard - though watch
out for some specific vendors which do not use the standard drivers.
M-Audio makes some nice hardware at good quality but can be
problematic for drivers, esp. if you have a 64-bit OS (XP64, Vista64
or linux amd64 based).
> So, I have 3 questions:
> 1. What is the source of this 60 Hz noise? Powerline? RF interference? I have tried disconnecting the power cord - it works sometimes, but not always.
Powerline, and/or badly shielded appliances, lights (flourecent) etc.
Watch out for ground loops if you have amplifiers in the line, beween
the microphone and the soundcard. Keep your lines short, use proper
twisted-cable balanced lines if possible, make sure there is a SINGLE
ground path (usually the desktop PC). Many laptops are NOT gounded.
Ofthen this is not a problem, otherwise make sure there is a single
Hum avoidance is a bit of an art, and requires careful
experimentation. Don't zap yourself!
> 2. Is there a standard way to get rid of the 60 Hz hum and ensure clean recordings for the purposes of speech recognition? The only way I can think of is to use notch filters at 60 Hz, and some of the harmonics, say 120 Hz and 180 Hz, since most of the times it doesn't manifest as a single tone at 60 Hz, but has harmonics all over the frequency range.
Many audio editing programs have hum removal tools, I would guess some
kind of adaptive filter should be best.
> 3. If this is a well-known problem with sound cards, then why aren't we, speech recognition community, trying to come up with a mandatory compliance standard for sound card manufacterers which ensures that the microphone input is not corrupted by 60/50 Hz noise? That should provide a big boost to speech recognition accuracies, and a wider acceptance of speech recognition.
The audio recording community has been on this for a long time. To
buy good equipment I would recommend going to a largish music store in
your area, and check their home-studio/small studio recording section.
Here at McGill in the speech coding lab, we use a M-Audio MobilePre
USB, connected with balanced cable to a AudioTechnica AT3035
microphone with pop filter, connected to a standard windows box. We
use Cooledit (no longer around) and Wavesurfer (free) to record, with
Joachim Thiemann :: http://www.tsp.ece.mcgill.ca/~jthiem