Transcript View our poster presentation! - clear

AUDIOFILES
Harika Basana ([email protected] ), Elizabeth Chan ([email protected] ), Nikolai Sinkov([email protected] ), Frank Zhang ([email protected] ) 6100 Main Street, Rice University, Houston, Texas 77005
GOAL
To explore the MP3 technology and
to implement various audio data compression
algorithms.
Analyze This
 Audio compression is to compress an audio
Ding.wav before compression
Ding.wav with frequencies within 1 std
from the mean
Original signal sampled at 44100Hz
After linear quantization
The x-axis DT sample and the y-axis is the amplitude
file into a smaller-sized file.
 People cannot differentiate between these
two files by just hearing.
 Due to its smaller size, the new file can be easily
transferred via the Internet.
 People try to find better audio compression
algorithms that retain satisfying audio quality.
Ding.wav with frequencies within 2 std
from the mean
Ding.wav with frequencies within 3 std
from the mean
Psycho Acoustic Algorithm
Algorithms
Linear, tangent or arctangent quantization of the
signal.
Average Energy Algorithm
Zeroes out selected high and low frequencies of
Procedure
 Perform the Discrete Cosine Transform (DCT)
the audio file.
After tangent quantization
After arctangent quantization
Masking Algorithm
The presence of a signal at a particular frequency can
raise the perceptual threshold of signals close to the
the masking frequency.
Procedure
 Go through every sample and remove the following
samples if they are below a certain threshold.
 Quantize the signal in one of the following ways :
Procedure
 Perform the Discrete Cosine Transform (DCT).
Diagram of the quantization “buckets” for the three
methods
Results
 No significant improvement. Need a better way of
implementing to get good results.
 Calculate the signal’s energy.
 Find the mean and the standard deviation of
from the energy spectrum.
Conclusion
 Keep all frequencies with energies within 1
standard deviation (std) from the mean.
 Used the underlying concepts.
 Zero out frequencies with energies outside this range.
 Similarly, keep frequencies with energies within 2
and 3 stds from the mean.
 We didn’t create MP3 files.
 Produced much smaller files.
 Give certain frequency bands more bits
(1000 – 5100 Hz and 12500 - 15200Hz).
 Throw away frequencies below 20Hz and above
20,000Hz.
 Perform the Inverse DCT and get the output.
 Perform the Inverse DCT.
Results
 Amount of compression is insignificant.
 Algorithm would probably work better if the signal
is very short, has monotonous tones, and has little
noise.
Results
 Compression is very significant.
 Quality is good for the amount of compression.
 Arctangent quantization yields the best quality.
 Psycho Acoustic Algorithm is the best, in terms of
- amount of compression
- sound quality of the output.
Improvements
 Implement windowing
 Implement temporal masking
Bibliography:
http://www.sospubs.co.uk/sos/may00/articles/mp3.htm
http://www.besar.dcs.gla.ac.uk/labs/audiolab/real_site/
tutorials/mp3/mp3how.php
and more…