Send the link below via email or IMCopy
Present to your audienceStart remote presentation
- Invited audience members will follow you as you navigate and present
- People invited to a presentation do not need a Prezi account
- This link expires 10 minutes after you close the presentation
- A maximum of 30 users can follow your presentation
- Learn more about this feature in our knowledge base article
APPLICATION OF SPEECH PROCESSING TO SWIFTLET SOUND
Transcript of APPLICATION OF SPEECH PROCESSING TO SWIFTLET SOUND
TO SWIFTLET SOUND
The framework analysis in speech processing applied toward swiftlet sounds
do not explored.
Swiftlet sound attraction recording evaluated by
human expert using try and error method
without specific analysis or synthesis about characteristic of the sound be causes swiftlet attract that sound.
At the end of the research,
we are achieved the objective our research which are:
implemented speech processing technique which are
MFCC and HMM
toward swiftlet sound.
analyzed three types of sound (
baby sound, adult sound and colony
) features use for swiftlet attraction in swiftlet husbandary premise.
applied the combination of feature extraction and classification and get classified percentage reached
Swiftlet industry becomes one of
in Malaysia based on potential and high demand on the bird nest for health care (Azman, 2011).
SITI NURZALIKHA ZAINI BT HUSNI ZAINI
PM Dr. Kamarul Hawari
PM Dr. Saiful Nizam
Previously, sound that produced at swiftlets husbandry premise actually is produced from
recording audio sound swiftlet voice without analysis
Within more this a decade, entrepreneurs
explored various methods and new technology
to increase production. Therefore, the research and development about sound of swiftlets attraction needed to technology develop swiftlets industry.
speech processing techniques toward swiftlet sounds.
the three different types of swiftlet sounds use in swiftlet house.
combination of feature extraction and recognition technique, due to classify the swiflet sounds.
Data sound for baby swiftlet, adult swiftlet and colony swiftlet we buy from
swiftlet farming industry, TCL Resources Sdn Bhd
where supply various swiftlet sound through Faculty of Science and Technology (FIST) Universiti Malaysia Pahang.
1. sample of sound
3. feature extraction
Sample of Sound
At frame 1,
features vector values about -36.94
features vector values -16.90
sound with -30.1 features vector values.
At frame 2
features vector value reached -4.72
features vector value -1.45
sound with 0.27 features vector value.
frame 2, frame 3, and frame 4
obviously shown the higher value start from baby swiftlet sound follwed by colony sound and the smaller value from adult sound.
frame 5 until frame 20
the feature vector values like nearest values for these three types of sound.
We can conclude that
same pattern at frame 2, frame 3 and frame 4 for three types of swiftlet sound because have increasing value.
But inverse at frame 1.
at frame 5 until 20 difficult to differentiate because the value only 0.10 feature vector values between this three sounds.
Take a sample of sound for 3 types of sound
Baby swiftlet sound
Adult swiftlet sound
Colony swiftlet sound
convert all sounds from .mp3 to .wav format, cut the sound into two seconds and filtering the environment sound
Extract features using Mel Frequency Cepstral Coefficient (MFCC)
Classify using Hidden Markov Model (HMM)
Matching the type of swiftlet sound
scope of research
There are three type of sound selected in this project for recognition at the end of research such as:
Produce by swiftlet baby using in internal house for make young bird comfortable in that house.
Produce by adult swiftlet mate to produce eggs in internal house.
Produce by a group of swiftlet using in puller house for call swiftlet fly in sky near and come to build their nest in swiftlet house.
Result & discussion
MFCCs of Baby Sound
MFCCs of Adult Sound
MFCCs of Colony Sound
This research get
Compare with (Clemins, 2003)
error (using MFCC and HMM) application on elephant sound.
That shown the MFCC and HMM reasonable for animals sound application not only apply in human speech.
From Robert et al. (2012) also
using same technique
(MFCC and HMM) for detection of bird species get best accuracy
and error only
Althought Robert et. al get higher than my research but the 88.7% still in acceptable in best accuracy range.
Desai et al. (2013) research from their experiment,
MFCC technique is superior to other techniques
when compared the result.
In animal speech application, Zeppelzauer (2005) stated that
MFCCs are well suited to discriminate the classes of animal sound.
That proven the use of the MFCC has the remarkable result in the field of speech recognition regard previous research. Therefore, MFCC was choosing in this research.
Klautau A. (2005) identified the
main step for MFCC
are pre-emphasis, framing, windowing, Discrete Fourier Transform (DFT), Mel-Filter Bank, Logarithm and Discrete Cosine Transform (DCT).
MFCC feature has been proposed
best popular extraction of speech
by Gaikwad et al. (2010) in their journal about speech recognition technique
There are two main stages in a speech recognition system, which are
training and testing stages.
Statistical representations of the temporal structure as well as spectral variation using Hidden Markov Models (HMMs) are the main technologies that have
contributed to the improvement of the recognition performances
Rabiner (1989) stated two strong reasons for the importance of HMMs. Firstly; HMM models are very rich in mathematical structure and hence can form the theoretical basis for
use in a wide range of applications.
Secondly, HMM models
work very well in practice
for several important applications provided that they are applied properly.
The main producers of edible nest in Asian country are
Birds produce sounds for
, with the majority falling in the categories of songs and calls.
Birds song to attract
mates or define territory
The animals generate sounds to
communicate with members of same species
Lee et al.
There are environmental factors such as
temperature, light intensity, humidity and sound is the key of successful place for swiftlets
because the swiftlet comfortable with environment like their
original habitat in caves.
swiftlets’s voice proven very effective attracts
swiftlets to be nested in bird house for swiftlets farming
the most interesting feature of swiftlets is
utilize a sonar-like system
to attract swiftlet.
Roger studied that swiftlets hearing responses to the frequency
1 - 16 kHz
and which most energy on
proven by James et al. (1993) in their paper.
The frequency also falls into human hearing (20 – 20 kHz).
Original habitat of swiftlet at the cave with
make human made swiftlet farming look like original habitat for swiftlet build their nest for industry.
Entrepreneurs make swiftlet house by consider environmental factors such as
aroma, light, temperature, humidity and sound
(Roger et al. 1987).
But, the swiftlets character is
sensitive toward sound
. Sound is the
to the swiftlets house (Mulia, 2007) .
Birds and humans produce sound of a
complex acoustic signal nature.
speech processing can be used for human and birds
Allinson and Patrica (1999)
Roger et. al