Birds Recognition using Pitch Frequency in Matlab
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
International Journal of Innovative Information Systems
& Technology Research 9(3):50-57, July-Sept., 2021
© SEAHI PUBLICATIONS, 2021 www.seahipaj.org ISSN: 2467-8562
Birds Recognition using Pitch Frequency in Matlab
Woko, Ovunda & Kabari, Ledisi Giok
Department of Computer Science,
Ignatius Ajuru University of Education, Port Harcourt, Rivers State, Nigeria
ABSTRACT
Birds sing, whistle or chirp. Whether they sing, whistle or chirp, it is of great importance for researchers
to extract the pitch frequency of their vocalization to deepen knowledge on their uniqueness. Pitch is an
important feature of birds’ vocalization which scientists can learn much about their distinguishing
inhabitants. To employ pitch as a feature, extraction method is fundamental to the researcher to have
accurate pitch estimation. The present technology of Digital Signal Processing plays important role in the
vocal signal extraction of bird’s pitches. This paper aims to use digital signal processing techniques in
Matlab to extract pitch frequencies of bird’s sounds to recognize their uniqueness. In this research, the
sounds of 13 different birds were collected from online website. The audio sounds of the birds were
recorded in 3GPP format accessible in MATLAB in WAV format. The names of the birds, their sounds
and duration of the record respectively are tabulated. The sampling frequency of collected data is 48,000
samples per second. We adopt Feature-driven development (FDD)- a form of Agile software development
methodology for this work. Matlab programming version R2018a tool was used for implementation. The
implementation follows our proposed model for pitch frequency detection. Time, frequency, and time-
frequency, power spectrum signal techniques are used. We observe and recognize the uniqueness of birds
by theirs pitch frequencies.
Keywords: Pitch, Frequency, Recognition, Digital Signal Processing, MATLAB
1. INTRODUCTION
There are about 10,000 to 13,000 bird species and 200 to 400 billion individual birds worldwide toady
(birdlist, 2021). However, the question about the number of bird species in the world depends on the
taxonomist. Identifying and classifying birds individually may not be a trivial task because no two birds
are the same. What feature can distinctly distinguish, detect and recognize birds? With advent of digital
signal processing using Matlab, the distinct detection and recognition of birds can be explored based on
distinct feature extract and analysis.
Bird pitch frequency recognition using digital signal processing involves feature extraction method,
feature classification and species detection (Briggs et al., 2012). Typically feature extraction method uses
time domain analysis, frequency domain analysis and time-frequency analysis to extract vocal features
from a bird data (Ghoraani, 2011). Feature extraction using a time domain analysis is very useful for
stationary data (Ghoraani, 2011; Vundavalli, 2016), however, it is not good because bird’s vocal data is
highly non-stationary. Feature extraction using frequency analysis is obliging for non-stationary signals as
well as stationary signal, and bird produces non stationary vocal signals. This analysis extraction method
is useful to a degree but due to the lack of temporal evaluation, the feature extraction is limited for non
stationary, whereas time-frequency analysis feature extraction method provide information about both
temporal and spectral features (Ghoraani, 2011; Starkhammar, 2015; Ghoraani, 2011). This reason leads
to selecting the time-frequency analysis for feature extraction.
50
Woko & Kabari ….. Int. J. Innovative Info. Systems & Tech. Res. 9 (3):50-57, 2021
Digital signal processing is a technique essential for a wide range of applications, from data science to
real-time embedded systems, and MATLAB products make it easy to use signal processing techniques to
explore and analyze time-series data, provide a unified workflow for the development of embedded
systems and streaming applications (Starkhammar, 2015). Exploring signal processing with MATLAB
products help one analyze signals from a range of data sources. We can acquire data, measure, transform,
filter, and visualize signals and apply signal processing tools to preprocess and filter signals prior to
analysis; explore and extract features for data analytics and machine learning applications; analyze trends
and discover patterns in signals; and visualize and analyze in time, frequency and time-frequency domain
characteristics of signals.
The analysis of birds vocal sounds have increased in the speech processing domain in the past few
decade. Much of the reported research has concentrated on the identification of bird species by
appearance. It is parochial to identify the bird variety directly by looking at it, because there are many
varieties of birds living around us. Trivial reported topic is the analysis of bird chirps from species of the
same bird. It is uncommon way to quantify difference between bird inhabitants by analyze the pitch
frequency of bird chirps or vocal sound because it require modeling, data acquisition, preprocessing,
extraction and pitch frequency recognition.
The paper presents a model for birds’ recognition using pitch frequency digital signal processing in
MATLAB to prove their uniqueness. We acquire the data set of vocal sounds of different birds. Develop a
model for processing the pitch. Import into MATLAB version R2018a and estimate the pitch frequency.
This work will greatly benefit zoologists ecological studies, researchers and socio-cultural community
about birds uniqueness by their pitch frequencies. .
2. Review of Related Work
The speech processing community in the last few years, has researched many issues in bird vocalizations,
especially species classification (Connor et al., 2012; Fagerlund , 2014; Trifa et al., 2008).
Babacan et al. (2013) evaluated pitch tracking on singing voice, with PRAAT and RAPT providing the
best determination of voicing boundaries.
Lachlan et al., (2013) developed Luscinia package widely used in the community to analyze bird song. It
provides pitch extraction but requires supervision. The Luscinia GUI allows the user to select elements
that require pitch tracking, but even when elements are carefully selected the exported pitch tracks may
not always match the spectrogram.
Meliza et al. (2013) developed Chirp, a tool which allows the user draw a mask on the spectrogram to
improve pitch estimation.
Hansson-Sandsten et al. (2011) analyzed male great reed warbler’s song based on its extracted syllables.
The analysis on the syllables extracted were done using time frequency analysis.
3. METHODOLOGY
3.1 Selection of Bird’s Sound
The data source of birds sound for this work was collected https://www.youtube.com/watch?v=WhRpW0
cVmds&t=143s. The birds’ sounds were downloaded, recorded, saved and used. The sound sources
acquired from the online sources should be laboratory generated or should be noiseless (Hansson-
Sandsten, et al., 2011) to avoid pollution feature extraction values. Noise removal algorithm may be
implemented to the sound acquired. Although denoising the sound with noise removal algorithms take
away the less amplitude feature extracted values and that may result in less efficiency and less accuracy
of the detection of the pitch.
In this research, the sound of 13 different birds were collected from online website (Kiddopedia, 2019) on
14th April, 2021. The audio sounds of the birds were recorded in 3GPP format and accessible in
MATLAB in WAV format. The table 1 shows the Birds’ names, sounds and duration respectively. The
sampling frequency of collected data is 48,000 samples per second. We adopt Feature-driven
development (FDD) - a form of Agile software development methodology for this work. Matlab
programming version 2018a tool is used for implementation.
51Woko & Kabari ….. Int. J. Innovative Info. Systems & Tech. Res. 9 (3):50-57, 2021
Table 1. Birds names, sounds and duration
No Name of Birds Chirping Duration Sound
in Seconds
1 Bald Eagle 5
Bald Eagle.3gpp
2 Blackbird 5
Blackbird.3gpp
3 Bluejay 5
Bluejay.3gpp
4 Bluetit 5
Bluetit.3gpp
5 Budgie 5
Budgie.3gpp
6 Crane 5
Crane.3gpp
7 Crow 5
Crow.3gpp
8 Eagle 5
Eagle.3gpp
9 Falcon 5
Falcon.3gpp
10 Nightingale 5
Nightingale.3gpp
11 Owl 5
Owl.3gpp
12 Pigeon 5
Pigeon.3gpp
13 Quail 5
Quail.3gpp
Data Source: from https://www.youtube.com/watch?v=WhRpW0cVmds&t=143s
52Woko & Kabari ….. Int. J. Innovative Info. Systems & Tech. Res. 9 (3):50-57, 2021
3.2 Pitch Detection Model
Our model collects birds sound, filters, processes time domain, frequency domain, power spectral feature
extract and correlate as shown in the figure1.
Bird sound Filter Time Frequency Power Feature Autoco
(Signal) Signal Domain Domain Spectral Extraction rrelatio
n
Signal 1 Signal 1 Signal 1 Signal 1 Signal 1 Signal 1 Autocorr
elation 1
Signal2 Signal 2 Signal 2 Signal 2 Signal 2 Signal 2 Autocorr
elation 2
. . . . . .
.
. . . . . .
.
. . . . . . .
. .
Signal n Signal n Signal n Signal n Signal n Signal n Autocorr
elation n
Figure 1: Pitch Detection Model for Birds
3.3 Detection of Bird Species mathematically
Auto-Correlation
Correlation is used to calculate the commonness between two variables. The correlation values will be in
between +1 and -1. When the correlation co-efficient is nearer to ±1, the two variables will have high
quantity of common elements. When the correlation co-efficient is nearer to 0, the commonness between
them is very low.
If the time series x(t) is correlated with another time series h(t).
The resultant correlation value will be r(τ ), that is r(τ) = x(t) * h(t) ……………………(1)
Where: t is the time series
τ is time lag
* is convolution factor
Correlation of a matrix or function with itself at different lags in time is called Auto-correlation.
If the time series x(t) is correlated with itself, then the resultant will be rx as
rx = x(t) ∗ x(t − τ ) …………………………………………..(2)
where: rx is the auto-correlation value
x(t − τ ) is delayed time series signal by τ
τ is time lag
(*) is convolution factor
53Woko & Kabari ….. Int. J. Innovative Info. Systems & Tech. Res. 9 (3):50-57, 2021
Applying auto-correlation to our data, means a violation the assumption of instance independence
because, the bird sounds recorded start at different instances. Therefore auto-correlation method helps to
rationalize and infer.
In our auto-correlation perfect multicollinearity is almost impossible. Perfect multicollinearity is getting
the auto-correlation value to exact ±1. Perfect multicollinearity is impossible. Though the sounds of a
same bird same but the electronic equipment used to reduce these sounds are different and also due to the
differences in specifications of components in different recorders, this condition occurs, even, when
source files are taken from the online database. The auto-correlated co-efficient must be nearer to ±1 in
order to say that the given audio signal is similar i.e. classified values are equal to a particular bird values
in our database. When the auto-correlated co-efficient is nearer to 0 (zero), the taken signal is not similar
to that of audio signal saved
4. RESULTS AND DISCUSSION
The results of the birds sound in table 1 are generated in time domain, frequency domain, power spectral
respectively. The implementation follows the proposed model.
Figure 2. Plot of Pitch Frequency of Bald Eagle in Frequency Domain
54Woko & Kabari ….. Int. J. Innovative Info. Systems & Tech. Res. 9 (3):50-57, 2021
Figure 3. Plot of Frequency normalization and Power Spectrum Bald Eagle
Figure 4. Plot of Pitch Frequency of Blackbird in Frequency Domain
55Woko & Kabari ….. Int. J. Innovative Info. Systems & Tech. Res. 9 (3):50-57, 2021
Figure 5. Frequency normalization and Power Spectrum of Blackbird
DISCUSSION
Closely observe the plots, the pitch frequency values of the birds sound are uniquely different. In the
power spectrum plot, Bald Eagle pitch frequency is about 0.13 while Blackbird is 0.06 dB as shown in
table 2. The implementation for the 13 different sample of birds gave different pitch frequencies but for
the sake of page consumption we captured, Bald Eagle are Blackbird. This shows that no two birds are the
same.
Table 2. Birds Pitch Frequency difference in Power Spectrum (dB)
Name of Birds Pitch Frequency in Power
Spectrum(dB)
Bald Eagle 0.13
Blackbird 0.06
CONCLUSION
The aim of this paper has been demonstrated by proposing a model for birds pitch frequency recognition
using digital signal processing techniques in Matlab to prove their uniqueness. The objectives are
achieved: data of vocal sounds of different birds were collected, model developed for processing pitch
frequency and power frequency spectrum implementation in MATLAB was carried. We have
demonstrated that birds can be recognition using pitch frequency using digital signal processing
techniques
.
REFERENCES
1. Birdlist (2021). List of Birds in the world. Retrieved 2021 from www.birdlist.org
1. Briggs, F. (2012). Acoustic classification of multiple simultaneous bird species: A multi- instance
multi-label approach. The Journal of the Acoustical Society of America, 131(6), pp. 4640–4650.
2. Ghoraani, B. and Krishnan, S. (2011). Time–frequency matrix feature extraction and classification
of environmental audio signals. IEEE transactions on audio, speech, and language processing,
19(7), pp. 2197–2209.
3. Vundavalli K. S. and Danthuluri V. S. R. S. (2016). Applied Signal Processing Blekinge Institute
of Technology SE–371 79 Karlskrona, Sweden
56Woko & Kabari ….. Int. J. Innovative Info. Systems & Tech. Res. 9 (3):50-57, 2021
4. Starkhammar, J. and Hansson-Sandsten M. (2015). Evaluation of seven time
frequency representation algorithms applied to broadband echolocation signals. Advances in
Acoustics and Vibration, vol. 2015
5. Connor, E. F. (2012). Automating identification of avian vocalizations using time- frequency
information extracted from the gabor transform. The Journal of the Acoustical Society of America,
132, 507–517
6. Fagerlund, S., and Laine, U. K. (2014). New parametric representations of bird sounds for
automatic classification. IEEE International Conference on Acoustics, Speech and Signal
Processing (ICASSP) (pp. 8247–8251). IEEE.
7. Trifa, V. M (2008). Automated species recognition of ant birds in a Mexican rainforest using
hidden markov models. The Journal of the Acoustical Society of America, 123, 2424–2431.
8. Babacan, O. (2013). A comparative study of pitch extraction algorithms on a large variety of
singing sounds. In 2013 IEEE International Conference on Acoustics, Speech and Signal
Processing (ICASSP) (pp. 7815–7819). IEEE.
9. Lachlan, R. F. (2013). The progressive loss of syntactical structure in bird song along an island
colonization chain. Current Biology, 23, 1896–1901.
10. Meliza, C. D. (2013). Pitch-and spectral-based dynamic time warping methods for comparing field
recordings of harmonic avian vocalizations. The Journal of the Acoustical Society of America, 134,
1407–1415.
11. Hansson-Sandsten M. (2011). Svd-based classification of bird singing in different time frequency
domains using multitapers,” in Signal Processing Conference, 2011 19th European, pp. 966–970,
IEEE, 2011.
12. Kiddopedia (2019). Birds Names and Sounds Learn Bird Species in English. Retrieved 20 21 fro
m https://www.youtube.com/watch?v=WhRpW0cVmds&t=143s
57You can also read
