Knn a machine learning approach to recognize a musical instrument

Gulhane Sushen .R, Shirbahadurkar Suresh .D, Badhe Sanjay, International Journal of Advance Research, Ideas and
Innovations in Technology.
© 2017, www.IJARIIT.com All Rights Reserved Page | 707
ISSN: 2454-132X
Impact factor: 4.295
(Volume 3, Issue 6)
Available online at www.ijariit.com
KNN- A Machine Learning Approach to Recognize a Musical
Instrument
Sushen R. Gulhane
D. Y. Patil College of Engineering, Pune,
Maharashtra
sushenrgulhane1@rediffmail.com
Dr. Suresh D. Shirbahadurkar
Zeal College of Engineering, Narhe,
Pune, Maharashtra
shirsd112@yahoo.in
Sanjay Badhe
D. Y. Patil College of Engineering, Pune,
Maharashtra
sanjay.badhe@dyptc.edu.in
Abstract: The integrated set of functions written in Matlab, dedicate to the extraction of audio tones of musical options connected
to timbre, tonality, rhythm or type. A study on feature analysis in today’s atmosphere, most of the musical information retrieval
algorithmic programs square measure matter based mostly algorithm so we have a tendency that cannot able to build a
classification of musical instruments. In most of the retrieval system, the classification is often done on the premise of term
frequencies and use of snippets in any documents. We have a tendency to gift MIR tool case, associate degree for recognition of
classical instruments, using machine learning techniques to select and evaluate features extracted from a number of different
feature schemes was described by Deng et al. The performance of Instrument recognition was checked using with different
feature selection and algorithms.
Keywords: Musical Instrument Recognition, Mel Frequency Cepstral Coefficient (MFCC), Fractional Fourier Transform
(FRFT), Machine Learning Technique (KNN).
1. INTRODUCTION
Recognizing the objects in the environment from the sound they produce is the primary function of the auditory system. The aim of
Musical instrument recognition is to identify the name and family of musical instrument from the sound they produce. Many attempts
were made for musical instrument identification and classification. The statistical pattern-recognition technique for classification of
some musical instrument tones with few features based on log-lag correlogram.
1.1Musical Instrument Classification
Depend on their shape, method of playing an instrument, the sound they produce: Musical Instruments are classified as follows:
An outline of the set of options which will be extracted with MIR tool cabinet, illustrated with the outline of 3 explicit musical options.
The tool cabinet additionally includes functions for applied math analysis, segmentation and bunch. The event of the tool cabinet is
to facilitate investigation of the relation between musical options and music-induced feeling.

Preliminary results show that the variance in feeling ratings may be explained by a tiny low set of acoustic options. We have a
tendency to select to base the planning of the tool cabinet on Matlab computing atmosphere because it offers sensible mental image
capabilities and offers access to an outsized type of different toolboxes. The various musical options extracted from the audio files
are extremely interdependent: particularly. Some options are supported same initial computations. So as to boost the procedure
potency, it's necessary to avoid redundant computations of those common elements. Every one of those inter-mediator elements, and
also the final musical options, are so thought of as building blocks which will be freely articulated one with one another. In several
existing audio retrieval system we have a tendency to extract the options either by linear prophetic code or by sensory activity linear
prediction. However in the projected system for extraction, we have a tendency to use musical data retrieval tool cabinet (MIR
toolbox) that is helpful to seek out audio descriptor by victimization hybrid choice methodology. Once finding audio descriptor we
have a tendency to establish the musical instruments with the assistance of vector division. Here we present an analytical framework
to determine the structure of digital music streams. In particular, we demonstrate techniques for music segmentation, segment
clustering, and summarization based on self-similarity analysis. Given structural features of digital music files, appropriate
information excerpts and “retrieval proxies” can help to automatically organize personal or commercial music collections. Our
methods depend on the analysis of a similarity matrix. The matrix contains the results of all possible pair-wise similarity comparisons
between time windows in the digital stream. The matrix is a better path to visualize and characterize the structure in digital media
streams. Throughout, the algorithms presented are unsupervised and contain minimal assumptions regarding the source stream. A key
advantage here is that the data is effectively used to model itself. The framework is intense general and should work on any ordered
media such as video or text as well as audio. In particular, we have demonstrated results on segmenting both speech audio 4 and video
streams 5 as well as music.
2. PROPOSED SYSTEM
In our proposed system, we are going to extract the features of sound by recognizing the timbre of the sound. After feature extraction,
we are making classification of sound on the basis of extracted features. For retrieving the audio data we use MIR tool box. MIR tool
box has the set of multiple functions written in Matlab. Those functions are used to extract the audio related features. In our proposed
system our main aim is to find out the audio descriptor. To find out an audio descriptor from given data we use extracted features and
hybrid selection method.
In hybrid selection method, we select the correct audio descriptors for the identification of singer of North Indian Classical Music.
Initially, only robust (primary) audio descriptors are released on the system in the first pass and its impact is noted. Then only select
the top few audio descriptors, having the largest impact on the identification process, are selected and remaining will be removed in
the backward or second pass.
This method reduces substantially the many numbers of audio descriptors to few specific audio descriptors. The selected audio
descriptors are then fed as input to further classifiers. After finding the correct audio descriptors we generate the feature vectors and
we will identify the musical instruments by using vector quantization method. In vector quantization system, feature vector stores the
extracted features of an audio descriptor and those extracted features will be matched with standard feature vector for comparison.

Training Phase
In this phase set of a known signal is used an input. The feature of the known signal will be extracted and this feature stored in a
matrix or vector format as a Reference Model which contain a standard database for identification.
Testing Phase
In this phase an unknown test signal will be given as an input and feature of the signal will be extracted. This feature can be compared
with the standard features. By using classifier we are able to recognize which feature matching among all feature. We are in a position
to recognize instrument and its family.
MIR Tool Box
MIR Toolbox, an integrated set of functions written in Matlab, dedicated to the extraction of audio files of musical features related,
among others, to timbre, tonality, rhythm or form. The objective is to offer a state of the art of computational approaches in the area
of Music Information Retrieval (MIR). The design is based on a modular framework: the different algorithms are decomposed into
stages, formalized using a minimal set of elementary mechanisms, and integrating different variants proposed by alternative
approaches – including new strategies.
We have developed that users can select and parametrize. These functions can adapt to a large area of objects as input. MIR
Toolbox is a Matlab toolbox dedicated to the extraction of musically related features in audio recordings. It has been designed in
particular with the objective of enabling the computation of a large range of features from databases of audio files that can be applied
to statistical analyses. We chose to base the design of the toolbox on Matlab computing environment, as it offers good visualization
capabilities and gives access to a large variety of other toolboxes. An analysis of the specific features considered in the
toolbox. All the different processes start from the audio signal and form a sequence of operations developed horizontally right wise.
The vertical composition of the processes indicates an increasing order of involution of the operations, from simplest measurement
to more detailed auditory modelling. Each musical feature is related to the different wide musical dimensions defined in music theory.
Machine Learning Technique (KNN)
KNN is non-parametric lazy learning algorithm. Non parametric means it does not make any assumption on the underlying data
distribution. Its outstanding characteristic is that it does not require a training stage in the strict sense. The training samples are rather
used directly by the classifier during the classification stage. The key idea behind this classifier is that, if we are given a test pattern
(unknown feature vector), x, we first detect its k-nearest neighbors in the training set and count how many of those belong to each
class. In the end, the feature vector is assigned to the class which has accumulated the highest number of neighbors. Therefore, for
the k-NN algorithm to operate, the following ingredients are required:
1. A dataset of labeled samples, i.e. a training set of feature vectors and respective class labels
2. An integer k ≥ 1.
3. A distance (dissimilarity) measure.

3. FLOWCHART
4. CONCLUSION
We have described a system that can listen to the musical instrument tone and recognize it. The work began with reviewing Blind
Source Separation and Musical Instrument Recognition. Features which make musical instrument distinct from each other are
presented and discussed. The principle of classifier k-NN is described.
Features are extracted from approximated sources and normalized to keep less complex. The k-NN classifier is used to
assess the testing data on this identification system. To make truly naturalistic evaluations, the acoustic data would be needed is
more.
From the above discussion we can say that, if we find the accuracy of identification with consideration of respective family
of an instrument, all feature provides more improved result than those are with instrument wise identification. The process becomes
very complicated if we try to find best feature value of an individual instrument.
REFERENCES
1. “Speech and Audio Processing” by Dr. Shaila D. Apte Professor, Rajarshi Shahu College of Engineering, Pune Maharashtra.
Formaly Assistant Professor, Walchand college of engineering, Sangli Maharashtra.M. Young, The Technical Writer’s
Handbook. Mill Valley, CA: University Science, 1989.
2. “Automatic musical instrument classification using fractional Fourier transform based- MFCC features and counter propagation
neural network” by D. G. Bhalke1 & C. B. Rama Rao1, & D. S. Bormane2 Received: 6 November 2014 /Revised: 16 April
2015 /Accepted: 16 April 2015 /Published online: 13 May 2015
3. # Springer Science+Business Media New York 2015.
4. “Automatic genre classification of Indian Tamil and western music using fractional MFCC” by Betsy Rajesh, D. G. Bhalke
Received: 14 January 2016 / Accepted: 6 June 2016_ Springer Science+Business Media New York 2016.
5. “Music Pitch Representation by Periodicity Measures Based on Combined Temporal and Spectral Representation”. Author:
PEETERS, G.
6. “Melody Extraction from Polyphonic Music Signal using STFT and Fanchirp Transform”, by Sridevi S. H. Department of
E&TC, DYPSOEA, Ambi, Talegaon, Pune, India. Prof. S. R. Gulhane, Department of E&TC, DYPCOE Ambi, Talegaon, Pune,
India

Knn a machine learning approach to recognize a musical instrument

More Related Content

What's hot (20)

Similar to Knn a machine learning approach to recognize a musical instrument (20)

Recently uploaded (20)

Knn a machine learning approach to recognize a musical instrument