Hi,
Could anyone tell me what is the criteria to switch between Sin and KBD mdct window? What is the advantage one over the other?
KWW
Full Version: window switching between sin and KBD window?
Hi, sorry a bit off-topic question. 
Are you making an AAC-implementation and if you are, is it possible you tell a bit more details about the project?
Are you making an AAC-implementation and if you are, is it possible you tell a bit more details about the project?
QUOTE
Originally posted by wkw
Hi,
Could anyone tell me what is the criteria to switch between Sin and KBD mdct window? What is the advantage one over the other?
KWW
Hi,
Could anyone tell me what is the criteria to switch between Sin and KBD mdct window? What is the advantage one over the other?
KWW
Search using Google for the string
"a single coefficient, the filter bank channels"
Skip to page 21.
--
Frank Klemm
I am just investigating the tools used in MPEG4-AAC focusing on the Gain Control tool. I am not sure how well the rest of the AAC tools can be integrated with the Gain Control filter bank, for example, the TNS, psychoacoustic models, LTP,MS-stereo, and also the limitation of the Gain Control tool. I only have a non MPEG compliant encoder and decoder.
I think the Sony Gain Control provides better transient detection and compensation together with Block Switching & TNS. The problem with switching to short-block is that it consumes too many bits. Long-block has higher coding efficiency than short-block and switching to short-block should be kept to the minimum.
Besides that, the Gain control tool can minimize pre-echo to a period of about 32 time samples or 0.7msec at 44100kHz(as stated in the MPEG4 standard) for both both block lengths. The original AAC mode uses block switching and TNS for pre-echo compensation, and the noise smearing of the short-block is rather high, (about 256 time samples, 5 msec)
I am still using block switching in the Gain control tool as I believe that the rest of the tools are not properly tuned this filter bank yet. For example, in the ISO reference src, the LTP I noticed doesn't have a ipqf filter bank to reconstruct the time domain samples needed for Long Term Prediction.. Does it mean that I can't use LTP with Gain Control tools?
Actually, the Gain Control technique belongs to Sony Corp. It is implemented in ATRAC-3. I think there is a website explaining about ATRAC-3.
wkw
I think the Sony Gain Control provides better transient detection and compensation together with Block Switching & TNS. The problem with switching to short-block is that it consumes too many bits. Long-block has higher coding efficiency than short-block and switching to short-block should be kept to the minimum.
Besides that, the Gain control tool can minimize pre-echo to a period of about 32 time samples or 0.7msec at 44100kHz(as stated in the MPEG4 standard) for both both block lengths. The original AAC mode uses block switching and TNS for pre-echo compensation, and the noise smearing of the short-block is rather high, (about 256 time samples, 5 msec)
I am still using block switching in the Gain control tool as I believe that the rest of the tools are not properly tuned this filter bank yet. For example, in the ISO reference src, the LTP I noticed doesn't have a ipqf filter bank to reconstruct the time domain samples needed for Long Term Prediction.. Does it mean that I can't use LTP with Gain Control tools?
Actually, the Gain Control technique belongs to Sony Corp. It is implemented in ATRAC-3. I think there is a website explaining about ATRAC-3.
wkw
QUOTE
Originally posted by Frank Klemm
Search using Google for the string
\"a single coefficient, the filter bank channels\"
Skip to page 21.
Search using Google for the string
\"a single coefficient, the filter bank channels\"
Skip to page 21.
(From the same article) For the sine window, DC energy is compacted in a single coeff. That's nice for image coding, but can anybody tell me when this is useful for audio coding?
Shouldn't you eliminate any (fixed) DC offset in an audio signal before encoding?
QUOTE
Originally posted by petracci
(From the same article) For the sine window, DC energy is compacted in a single coeff. That's nice for image coding, but can anybody tell me when this is useful for audio coding?
Shouldn't you eliminate any (fixed) DC offset in an audio signal before encoding?
(From the same article) For the sine window, DC energy is compacted in a single coeff. That's nice for image coding, but can anybody tell me when this is useful for audio coding?
Shouldn't you eliminate any (fixed) DC offset in an audio signal before encoding?
There's no easy way to remove DC signals. Although MPEG suggests to use
a highpass (2...10 Hz) before encoding I strongly unrecommend this.
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.