Hi,
Why tonal signals have less masking capability compared to noise signals.
Regards
Shreya
Full Version: why Tonal signals have less masking capability
QUOTE(shreya_pathak @ Oct 19 2004, 01:31 AM)
Hi,
Why tonal signals have less masking capability compared to noise signals.
Regards
Shreya
Why tonal signals have less masking capability compared to noise signals.
Regards
Shreya
I don't know why.. perhaps that is why the Tone Masking Noise, TMN value is as high as 18 dB in AAC and 29 dB in MP3 (taking into account bin-aural masking level differences ).. A bigger TMN value means a greater "unmasking" at a particular partition band..
wkwai
You could consider noise as the sum of many tones.
In fact some psy models (mainly from JJ and F. Baumgarte if I remember well) are not even considering tonality. They consider every spike as a masker and add them. This also leads to more masking in noisy areas, without computing any tonality value.
In fact some psy models (mainly from JJ and F. Baumgarte if I remember well) are not even considering tonality. They consider every spike as a masker and add them. This also leads to more masking in noisy areas, without computing any tonality value.
QUOTE(Gabriel @ Oct 20 2004, 09:36 AM)
You could consider noise as the sum of many tones.
In fact some psy models (mainly from JJ and F. Baumgarte if I remember well) are not even considering tonality. They consider every spike as a masker and add them. This also leads to more masking in noisy areas, without computing any tonality value.
In fact some psy models (mainly from JJ and F. Baumgarte if I remember well) are not even considering tonality. They consider every spike as a masker and add them. This also leads to more masking in noisy areas, without computing any tonality value.
How do you compute a tonality value? Peak^2 / Total energy in some frequency band around it? Other methods?
QUOTE(Gabriel @ Oct 19 2004, 11:36 PM)
You could consider noise as the sum of many tones.
In fact some psy models (mainly from JJ and F. Baumgarte if I remember well) are not even considering tonality. They consider every spike as a masker and add them. This also leads to more masking in noisy areas, without computing any tonality value.
In fact some psy models (mainly from JJ and F. Baumgarte if I remember well) are not even considering tonality. They consider every spike as a masker and add them. This also leads to more masking in noisy areas, without computing any tonality value.
Noise as the sum of many tones ??
Actually, this is a very hot topic in psychoacoustic discussion, since there is not a definite conclusion - I'll point out one which looks most logical to me (but beware - there are others)
There are two things:
(1) Masking power in a critical band (frequency)
- Tone is usually an isolated masker (peak) - i.e. single masker
- Noise is, in most cases, plurality of individual maskers (flat spectrum) - i.e. many maskers
(2) Masking power in time
Human nervous system is particulary sensitive to uncertanity in temporal domain - tonal signals have no uncertanity while noise signals have very high uncertanity - this also attributes to high masking power of noise components.
There are several mathematical approaches to model this phenomenon - two of which are used by ISO as an example - spectral flatness measure (SFM) which only takes (1) into account and euclidian distance between predicted and actual signal in the future which partially takes into account (2)
Also, there are alternatives - power sum of individual maskers as mentioned Gabriel (and used by ITU-R BS.1387), temporal uncertanity measure by JJ, and physiological ear model which I didn't study too well to comment how this attribute is modeled.
There are two things:
(1) Masking power in a critical band (frequency)
- Tone is usually an isolated masker (peak) - i.e. single masker
- Noise is, in most cases, plurality of individual maskers (flat spectrum) - i.e. many maskers
(2) Masking power in time
Human nervous system is particulary sensitive to uncertanity in temporal domain - tonal signals have no uncertanity while noise signals have very high uncertanity - this also attributes to high masking power of noise components.
There are several mathematical approaches to model this phenomenon - two of which are used by ISO as an example - spectral flatness measure (SFM) which only takes (1) into account and euclidian distance between predicted and actual signal in the future which partially takes into account (2)
Also, there are alternatives - power sum of individual maskers as mentioned Gabriel (and used by ITU-R BS.1387), temporal uncertanity measure by JJ, and physiological ear model which I didn't study too well to comment how this attribute is modeled.
QUOTE(Ivan Dimkovic @ Oct 21 2004, 11:18 AM)
Actually, this is a very hot topic in psychoacoustic discussion, since there is not a definite conclusion - I'll point out one which looks most logical to me (but beware - there are others)........
And which one is used by Nero? :-))
Sorry, but I am afraid that this answer falls into the "trade secret" category 
I'd guess peak detection (SFM)
note: just a personnal guess without any insight knowledge from Ahead's audio encoder
note: just a personnal guess without any insight knowledge from Ahead's audio encoder
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.