AAC concept doubts ! explanation needed

[Dave1024]

Hi,

I would like to know the following core concepts relate to the audio compression algorithms AAC

1. What is mean by a window? and what is the relation of a window with respect to the sampling frequency.

2. What is mean by window overlapping?

3. What is the use of Temporal Noise Shaping (TNS)?

4. When spectral band replication is used what is the criteria used for finding the treshold frequency?

5. What is the difference between joint stereo, mid side stereo and parametric stereo ? how the encoder take a decision of method ?

6. What are the popular post and pre processing in audio like in loop filter in video.

Please give some reference spots for the above doubts to go in depth in to the AAC standard.

:thanks:
Dave

[Mug Funky]

this is as i understand it... i'm not an audio codec coder :)

1. What is mean by a window? and what is the relation of a window with respect to the sampling frequency.

window is analogous to macroblocks in video coding. the input PCM stream is split into blocks which are called windows because (i presume) of it's similarity to "sliding window" types of frequency transforms.


2. What is mean by window overlapping?

the windows overlap 50%, so any edge errors are "crossfaded" into each other between windows and thus cancelled out.

consider it analogous to overlapping-block techniques in video.

3. What is the use of Temporal Noise Shaping (TNS)?

from http://www.mp3-tech.org/aac.html

"Temporal Noise Shaping (TNS) is a tool designed to control the location, in time, of the quantization noise by transmission of filtering coefficients "

sounds like it's similar in concept to trellis and psy-RD in video.

4. When spectral band replication is used what is the criteria used for finding the treshold frequency?

no idea. probably encoder-specific. IIRC SBR is proprietary and made it's way into the standard a little later. most times i've encountered SBR techniques, the encode simply works off a resampled original (like 22.05k core with SBR bumping it up to 44.1k)

5. What is the difference between joint stereo, mid side stereo and parametric stereo ? how the encoder take a decision of method ?

joint is a generic term for any kind of stereo encoding that isn't discrete LR. mid-side is L+R and L-R matrixed, so more bits can be spent on L+R (the "middle" mono part), and less on the L-R part (the "sides" of the stereo image) where this would provide an efficiency gain. theoretically if the left and right channels are identical (mono-in-stereo), MS stereo will use half the bitrate for the same result.

i've no idea about parametric stereo.

6. What are the popular post and pre processing in audio like in loop filter in video.

i would probably avoid preprocessing - audio tools are not as sophisticated as video tools (audio is less dimensional than video and so it's more difficult for computers to differentiate certain elements as well as a brain can).

psy-models in lossy audio are geared toward natural audio. any processing serves to make the audio less natural. compare a 2010 pop recording's performance in VBR to one from 1980 and you'll see a sharp increase in bitrate used.

[shon3i]

5. What is the difference between joint stereo, mid side stereo and parametric stereo ? how the encoder take a decision of method ?In addition to Mug Funky, parametric stereo or PS is technique that encode stereo stream as mono but wrote information about stereo image. Decoder must support PS decoding otherwise stream will be decoded as mono, using that information decoder should reconstruct near to original stereo image. PS should be used on very very low bitrates it's realy good technique to make decent results with 32kbps and lover.

[Dark Shikari]

3. What is the use of Temporal Noise Shaping (TNS)?

from http://www.mp3-tech.org/aac.html

"Temporal Noise Shaping (TNS) is a tool designed to control the location, in time, of the quantization noise by transmission of filtering coefficients "

sounds like it's similar in concept to trellis and psy-RD in video.It's moreso a way of eliminating pre-echo by reducing the amplitude of ringing artifacts.

4. When spectral band replication is used what is the criteria used for finding the treshold frequency?

no idea. probably encoder-specific. IIRC SBR is proprietary and made it's way into the standard a little later. most times i've encountered SBR techniques, the encode simply works off a resampled original (like 22.05k core with SBR bumping it up to 44.1k)If it's in an open standard, it's not proprietary. SBR has no threshold frequency: it's simply a doubling of frequency.

5. What is the difference between joint stereo, mid side stereo and parametric stereo ? how the encoder take a decision of method ?Parametric stereo is a method of approximating stereo using a filterbank instead of coding two separate channels.