Re: [Audacity-quality] Tests of new Noise Reduction effect

[Bill W. <bi...@go...>]

On 15/11/2014, at 3:44 PM, Bill Wharrie <bi...@go...> wrote:

> [snip]
> 
> Preliminary results
> 1) The new NR compares favourably to DeNoise (which has been my “gold standard”). This is with default settings for the new controls, and the “old” controls set to 0/0/0/0 (Sensitivity/Frequency Smoothing/Attack/Release). The new NR seems to achieve a greater perceived NR than DeNoise, due, I think, to the fact the DeNoise does not apply downward-expansion to bands “nearby” a band that is determined to contain signal. You can see this “nearby” effect in a tone+noise test.

Wrong!

Comparing two identical tracks, one processed with DeNoise and the other with Audacity, both asking for 24 dB noise reduction, DeNoise seems to apply 16 dB reduction. I measured this using Plot Spectrum on white noise. I don’t know if this is a bug in DeNoise, or if DeNoise is using some other measure of “noise floor”. In any case I’ll be careful in the future to make sure I’m doing fair comparisons between DeNoise and Audacity.

— Bill