3. INTRODUCTION
3/21
There are over nine thousand bird species !
Goals in this work is to develop methodology for the
system that could automatically recognize bird species
or even individual birds by their sounds
4. WHY IT IS IMPORTANT
4/21
Monitoring birds by their sound is important for many
environmental and scientific purposes.
Like,
- reduce the need of volunteers in this kind biological
project
- observed easily by experienced bird watchers
- identify and count birds in a specific area
- to estimate long-term population trends
5. LITERATURE REVIEW
5/21
Automatic classification of bird species from bird sound
samples has recently attracted the interest of the research
community because of the improvement of different techniques
in
• signal processing and
• machine learning
6. LITERATURE REVIEW
6/21
Signal processing is a broad term that involves the
use of audio processing techniques to improve
signal quality and extract a set of features from the
audio signal
8. LITERATURE REVIEW
8/21
Birds produce their sounds mainly
by syrinx, which is unique organ for
birds.
-sounds can be broadly classified
as songs and calls
-which can be further divided into
hierarchical levels of phrases,
syllables and elements or
notes.
9. LITERATURE REVIEW
9/21
-songs are longer vocalizations which usually include a
variety of notes in a sequence
-while bird calls are short communications which are
often the single notes
11. LITERATURE REVIEW
11/21
Used technique for bird sound processing:
- energy based time-domain approach
which is reliable for single bird’s samples with low
noise
- multiple bird’s sounds in noisy environments,
two dimensional time–frequency based
segmentation is used
12. LITERATURE REVIEW
12/21
The most widely used features to describe bird’s sounds are
- Linear predictive coefficients (LPC) and
- Mel-frequency cepstral coefficients (MFCCs)
13. LITERATURE REVIEW
13/21
Classification methods
that were mostly used for training/testing the data-set
- K-Nearest-Neighbour (k-NN)
- Naive Bayes
- Support vector machines
- Random forest
- J-48 decision tree
- Neural networks
15. METHOD
15/21
A. Collection of Database and Pre-Processing of Audio
Wave
-Noise reduction techniques (like Butterworth filteris) have to be used to remove
from signal some unwanted noise components
like, wind, rain etc
16. METHOD
16/21
B. Feature Extraction
Mel Frequency Cepstral Coefficients(MFCCs) are the most used features used to
describe the spectrum of an audio recording in very compact yet informative
manner.
19. RESEARCH SCOPES
19/21
Performance of the different proposed bird
identification system still leaves scope for
improvement.
• to use more data
• audio recordings in noisy environments with
multiple bird species simultaneously
• find more efficient and robust classification
techniques and features to improve classification
performance
• real-time audio recording and process
