reported
by: Dr. Roland K C Tan
Section Advisor
(Term 1999/2000)
On Friday, 17 September 1999
at 7.00 pm, 6 members and 24 guests of the AES
Singapore Section gathered at the Lecture Theatre 59, Ngee Ann
Polytechnic, for some light snacks and refreshments before a seminar
presentation by Mr. Cedric M.M. Tio, an AES Student Member.
Cedric is currently a postgraduate student at the Centre for High
Performance Embedded Systems (CHIPES) at Nanyang Technological University (NTU).
He was my undergraduate project student between 1997 to 1999 while he was
pursuing his Computer Engineering degree at NTU.
The presentation in the evening
was based on our paper we have submitted for the AES 107th Convention and
was a culmination of Cedric's final- and honour-year project work at NTU
where he would be presenting it on 27 September 1999 in New York, USA. His
trip to New York was kindly sponsored by Creative Technologies which also
partially sponsored his honour-year undergraduate studies in 1998/99.
At around 7.30pm, I first gave a
short presentation of an overview of the AES and what we stood for - our
mission as well as our past years' activities and achievements, on behalf of
the Section Chairman, Mr. Jibby Jacob. He was unable to attend this
seminar as he had to attend his company annual dinner & dance where he would
be receiving his long service award from the Radio Corporation of Singapore
(RCS) - a company he has worked for over 15 years.
At 8.00 pm, Cedric began
his talk on MASC by first giving the audience a short sound demonstration by
comparing the results of his work with some original CD recordings. This was
followed by a brief overview of the proposed MASC coding algorithm which
include a brief discussion of the QMF filter bank design and MLPC algorithm.
MASC, or the multi-pulse adaptive sub-band coding algorithm, proposed
earlier when I was a PhD student under Prof. Malcolm Hawksford at the
University of Essex in 1993, is a hybrid source coding algorithm that
combines multi-pulse linear predictive coding (MLPC) with sub-band coding
(SBC) for digital audio applications.
|
Cedric Tio
discussing on the proposed MASC parametric coding technique for digital
audio applications. - photograph by Robert
Soo. |
In MLPC, signals are not
classified into voiced or unvoiced frames. Instead the excitation consists
of a number of impulses for each frame so that the final result is a closer
approximation to the original PCM signal which makes it suitable for
high-quality audio applications. The coding performance of MLPC is sensitive
to the number of pulses placed in each frame of input samples. For broadband
audio signals, more excitation pulses are required in each frame in order to
model the waveform envelope accurately. It was shown in that the performance
of MLPC can be further improved by dividing the broadband audio signal into
narrower sub-bands using SBC technique. The substantial improvement is
brought about by the simplification of the time-domain waveform signal as a
result of the narrow band filtering process.
In the earlier MASC design, a
bit-rate reduction ratio of 7:1 per channel was achieved with promising
results using 8-band. Cedric has done further investigations on the
effects of a high-order sub-band filter bank MASC system by extending the
number of bands to 32 using quadrature mirror filter (QMF) bank. The MPEG
psychoacoustic model was incorporated into the MASC algorithm to further
enhance the performance of its sound quality. Based on this psychoacoustic
model, a new dynamic excitation pulse allocation algorithm was developed.
The proposed new pulse allocation
algorithm consider auditory masking by distributing the total number of
pulses to each sub-band according to their signal-to-mask ratio (SMR)
values. The psychoacoustic enhancements make efficient use of the limited
available excitation pulse pool. Heavily masked sub-band signals, which have
lower SMR values, are coded with a relatively lower number of excitation
pulses. The pulses saved can be allocated to encode other sub-bands with
higher SMR values. The correlation between the number of excitation pulses
and the SMR values is evident in the obvious improvements in the perceptual
quality of audio samples coded with and without the psychoacoustic model.
This has been verified earlier during the sound demonstrations.
According to Cedric, the
increase in coding efficiency brought about by the introduction of the
psychoacoustic model and the QMF filter banks was, however, compromised by
the effects of the high decimation process. The process of a high decimation
process at a factor of 32 on a narrow sub-band signal resulted in a
broadband signal. This is not suitable for use with LPC which works better
with a narrower band signal. It resulted in a much poorer overall bit-rate
reduction ratio than expected from the initial figure that was achieved in
the earlier MASC design at 7:1. In his conclusion, an 8-band MASC system was
proposed together with the new dynamic excitation pulse allocation algorithm
developed.
During the Q&A session, questions
were asked about whether or not MASC has considered the adaptation of the
Fletcher & Munson curve in its psychoacoustic model computation. Cedric
explained that the absolute threshold curve incorporated in the
psychoacoustic model of the MPEG audio encoder, and which was also adopted
in the MASC algorithm, was actually based on the Fletcher & Munson curve.
Alternatively, an F-weighting function curve could also be incorporated.
Another member of the audience also suggested exploring the possibility of
using similar pulse allocation sequence in the various sub-bands that could
be highly correlated to further reduce the overall bit-rate.
|
Mr. Toh Chee
Weng (right), AES Student Member and NTU graduate student, receiving a
letter of invitation from Mr. Foo Jek Luan to attend a Yamaha pro-audio
demo at Carnegie Hall in New York. -
photograph by Robert Soo. |
Mr. Cedric Tio
(right), AES Student Member and NTU graduate student, receiving a letter
of invitation from Mr. Foo Jek Luan to attend a Yamaha pro-audio demo at
Carnegie Hall in New York. - photograph by
Robert Soo. |
| At around 9.00pm
after the Q&A session, this was followed by a short speech delivered by
Mr. Foo Jek Luan, Managing Director of Audio-Visual Land Pte Ltd,
before a short demonstration of some Yamaha pro-audio equipment. The
Singapore based company is a local distributor for Yahata pro-audio
equipment. In his speech, Foo voiced his support for the AES
Singapore Section and look forward to working more closely with the
local section in the future. The company had also signed up to become a
Sustaining Member of the AES before the meeting ended at around 9.30pm. |
Mr. Foo Jek
Luan (left), Managing Director of A-V Lands Pte Ltd, handling over a
bank cheque of US$750 and the AES Sustaining Membership application form
to Dr. Roland Tan who was representing the Singapore Section Chairman. -
photograph by Robert Soo. |
|
A demonstration
of the Yamaha digital pro-audio equipment by A-V Lands Pte Ltd. -
photograph by Robert Soo. |
The AES
Singapore Section would like to thank their fellow committee member,
Mr. Rafael Oei, Lecturer at Ngee Ann Polytechnic, for the use of the
lecture theatre and its AV facilities. |
Biography of Mr. Cedric Tio
Mr. Cedric Tio,
Student Member, AES Singapore Section,
MEng Student, CHiPES, NTU, is
currently based at the Centre for High Performance Embedded Systems (CHiPES),
Nanyang Technological University in Singapore. He is near the completion of
a Masters Degree by research in Computer Engineering, specializing in
digital audio signal processing. He has done some extensive research work in
audio data compression and audio watermarking. His background in computer
science allows him to dabble in software programming and IT programming once
in a while. He is also an amatuer musician who plays the guitar, piano and
drums. In his free time, he composes and record music for a commercial
animation studio. |
