MPC - MP2

MPC is only distantly related with the ancient MP2 encoders.
It's a completely self written encoder, based on subband compression which makes use of filterbanks.

There are many difference between MPC and MP2 codecs.
MPC introduces the following features:

- VBR
- M/S Coding
- ANS (Adaptive Noise Shaping)
- Huffman (lossless compression) but the main difference is the psychoacoustic model with its extensions.

"- because the resolution in freq.domain is much less than the resolution with transform codecs, mpc uses so-called "adaptive noise shaping" to do a spectral shaping of the quantization noise in the subbands. This proceeding saves bitrate especially within the lower subbands".

- to get a much higher lossless compression, mpc uses
huffman-coding and more sophisticated coding-schemes like differential coding.


psychoacoustics (from old Buschmann's post but still good):
=============
also the psychoacoustic model has been tuned a lot. it was based on the "psychoacoustic model 2" that was introduced with the MPEG1-standard. But i implemented a lot of new features like "clear voice detection" (cvd), exploitation of temporal postmasking, detection of fast transient signals, nonlinear spreading-function, new threshold in quiet and so on... also the
psychoacoustic model takes filterbank-specific attributes (e.g. aliasing) into consideration.
CVD is able to detect voice-like signals to give a higher quality with voices or sounds with harmonic spectra. it uses special analysis to detect harmonics with varying base frequency - the "normal" psychoacoustics are not able to
detect such signals and will add audible noise to these signals.
The exploitation of temporal postmasking saves a few kbit/s because the human hearing has to "relax" after a sound event, so that the encoder can put a bit more distortion to the signal during this time.
Another thing that had to be changed was the threshold in quiet.
The threshold in quiet is the loudness below noone can perceive a sine-tone or a narrowband-noise. The threshold in quiet (ltq: "l"evel "t"reshold in "q"uiet) described in the ISO-MPEG-standard is the ltq that was measured with average listeners (me, my mom, my grandma and so on...), so that also
listeners were included that have not the best hearing. the new ltq was measured by listening tests of three (mine and my major listening testers) persons using very precise headphones.
The results of the psychoacoustics are much better than with the use of the iso-ltq.
Another change in the psychoacoustics is the use of nonlinear spreading-function. In fact the masking effect the spreading-function describes is nonlinear - it varies with the frequency and the loudness of the masker.
The "normal" psychoacoustic model does not exploit this fact.
===============

    ritorna alla pagina Principale

                          ritorna alla pagina Muse PaCk