How good is sigma-studio and ADAU1446?

Hi Russ,

SigmaDSP and SigmaStudio were originally conceived for simple audio post-processing (EQ, etc.). Since then it has grown to perform some impressive things. In one of our products a lowly ADAU1701 replaced several dedicated chips while doing a better job. Searching this forum reveals many other examples of the serious work one can do with it. That being said, your application (in fact, any such that attempt to attenuate or cancel noise, which my Filters 101 professor termed The Enemy) is rather difficult. It may lie beyond what SigmaDSP can do for you. Let's examine the parameters of your application before considering some alternatives:

I assume you're already using a quality super-cardioid microphone. Anything you can do to improve the S/N going in will directly effect an improvement..

Performance of the frequency-selective noise gate improves with the size of its filter bank. If you have the instructions available, going to six or more bands per octave will help -- since the filters can then be more selective (with higher Q). Unfortunately, building one of these is a bit tedious, especially since SigmaStudio gets a bit sluggish when editing truly huge projects. Also be sure to save your work every half-hour or so. Ultimately,though, success of a huge frequency-selective noise gate goes back to the S/N from your mic. As the incoming S/N diminishes, more filters trigger on noise instead of signal. Too much of that, and your system is -- well, doomed.

You're competing with the human ears and brain, which are already wonderfully efficient at pulling signal out of noise. Commonly known as the "cocktail party effect", it allows most people to understand speech barely above ambient noise. In fact, when someone speaks our own name, often we can recognize it at levels several dB below the noise! A recording made from the same location does not confer this ability, showing that a mic can't compete with two ears. Also, the inner ear is a marvel of engineering -- with a frequency-selective canal lined with hairs that either get triggered or not, it amounts to a digital Fourier spectrum analyzer!

This last fact provides a clue toward better noise-reducing applications. Some of the neatest stuff in DSP -- including such things as MP3 encoding -- are done in the frequency domain. SigmaDSP only recently acquired this capability in the ADAU145x processors, but so far, SigmaStudio only supports very basic frequency-domain processing. I don't know about more sophisticated DSP, but perhaps you can ask around in the SHARC forum about better noise-reducing methods.

Best regards,

Bob

Hi Bob,

Thank you very much for your detailed response. Our previous systems use a unidirectional entertainment microphone. We will probably upgraded the microphone and the PA system. I have been doing some research about 'spectral subtraction' and this seems to be the most recommended route to follow, do you agree? But there also seems to be several drawbacks. Particularly when you calibrate the microphone i.e. when you measure the frequency of the voice, and also that of the noise (ie. what you want to subtract). I have been in contact with another company to see what devices that might be able to offer but I was really just double checking that the route of using a dual microphone that will subtract background noise is the best way forward.

Thanks

Russell