Automatic Notes Generation for Musical Instrument Tabla

Transcription

1 Volume-5, Issue-5, October-2015 International Journal of Engineering and Management Research Page Number: Automatic Notes Generation for Musical Instrument Tabla Prashant Kanade 1, Bhavesh Chachra 2, Harmeet Sawlani 3, Vishal Jumnani 4, Deepak Bhagtani 5 1 Assistant Professor, Department of Computer, V.E.S.I.T, INDIA 2,3,4,5 Student, Department of Computer, V.E.S.I.T, INDIA ABSTRACT In this paper, our purpose is to design an intellectual system which will generate the notes of the tabla based on the input string provided by the user. The main idea of the proposed system is divided into two phases the string checking and generation phase and the string playing and storage phase. This system will accept the strings and check whether the input strings are valid or not and will apply some probability distribution based algorithm on strings and generate some combination of strings (tones generation from string) in the string checking and generation phase and the generated combination of strings (generated tones) which are melodious are played and the tone which is selected is saved in the users database and downloaded in computer. Keywords---- Teen, Rhythm, Western music I. INTRODUCTION Indian classical music is one of the most oldest and unbroken traditions in the world. It s origin stem from the Vedas, the ancient script of the Hindus. Presently, Indian classical music is based upon two main elements: Rag and Tal. Rag is the melody of the music while the Tal is concerned with the rhythm. This project aims to deal with the idea of rhythm in Indian classical music, specifically the control of rhythmic patterns using an instrument called the Tabla. Tal literally means clap. It implies the control of rhythmic patterns in Indian classical music by the use of hand claps. Today the Tabla is used in place of the clap in a performance but the origin of the word remains. The Tabla consist of 2 drums the Dayan and the Bayan. The instrument is used not to just control the rhythmic patterns but can also create large, complex compositions based upon rhythmic structures and strict formalisms. One of the most unique attributes of Indian classical music is the oral transmission of music between musicians and listeners due to the lack of formal written notation unlike Western music. As a result Tabla compositions bare a strong relationship to linguistics and natural language. It is my intention to explore this idea by trying to construct a program which uses formal grammars taken from the study of linguistics as a way in which to model Tabla compositions. This fundamentally implies that Tabla compositions work as a language. David Courtney, an expert Tabla player has made an analogy about learning the theory of Tabla which supports the connection between Tabla compositions and linguistics in Just as knowledge of grammar separates a literate speaker from the masses, likewise an understanding of the theory of Indian rhythm and compositional form separates a master from one who is merely functional. II. DESCRIPTION Kaida: Kaida are basically rules which are used to specify the bols present in the tabla. There are different types of kaida. Tabla: Tabla is an ancient Indian musical instrument; it is used by classical music artists. The tones of tabla are called Alphabets. There are about bols. Each bol represents the sound which is generated by playing tabla. Each stroke played on tabla represents sound which can be vocally pronunced. Each bol is specific. Combination of these bols is used to play one string. Long and long patterns can be generated using these bols. While constructing these patterns two things are to be considered Raag & Taal. Raag is melody of the pattern generated and Taal is the rhythm of the words.the Tabla has pair of drums which has right-hand drum Dayan & left-hand drum Bayan. Talis nothing but "clap". The clap used in the performance is replaced by tal. Tali or bhari, khali, bol, theka, avartan, vibhag or (ang), lay, sam & matra. Certain rules are to be followed while constructing the patterns, some of the rules are Kaida, Teen tal (16 beats). For instance teen tal can be shown as: Dha Dhin Dhin Dha Dha Dhin Dhin Dha Na Tin Tin Ta tete Dhin Dhin Dha 326 Copyright Vandana Publications. All Rights Reserved.

2 1. N-gram Modeling III. METHODOLOGY USED N-Gram modeling is an approach of data analysis designed for sequential streams of data. The principle is elementary, quite literally. A stream of data is split into its constituent terms, or n- grams, as they are called, and the information extracted from the relationships between individual n-grams is used to construct an n- gram model, which can then be used for tasks like predicting the next term in a sequence, or constructing new sequences similar to an existing one. The concept is easily demonstrated using a simple example. Let us consider a sequence of letters S = {ABAACAABAA}. Let us construct an n-gram model on this data using a simple histogram. Splitting the stream into its constituent letters, we see that there are 7 As, 2 Bs and 1 C. We can now estimate P (A) = 7/10, i.e. there is a 70% chance that the next element in the sequence will be an A. A more sophisticated approach would be to use bi-grams, or groups of 2 letters. To estimate the probability of the next term in the sequence being an A, we are looking for the probability of the bi-gram AA, since the last term in the sequence S is an A. P(A A) = P(AA)/(P(AA) +P(AB) +P(AC)) = 3/( ) = 50%. This means that the probability of A being the next element in the sequence is 50% and not 70% as we predicted earlier! Histogram showing the list of all possible bi-grams and their counts in a sequence S = {ABAACAABAA}. Bi-gram AA AB AC BA BB BC CA CB CC Counts A model that uses still more information can be constructed using sequences of tri-grams, or even 4-grams and so on. N-gram modeling has proved itself to be a robust approach to sequential data analysis. It has been applied rather successfully in speech processing and statistical natural language processing, and is now being applied to music modeling as well. Perhaps the most important advantage of using n-gram modeling is that it has an inherent approach to addressing the hierarchy within a sequence. In music, as with language, a single stream of events can often be interpreted in multiple ways - as a stream of notes, as a set of gestures and short phrases, or as a coherent melody - and considering n-grams of different lengths allows us to consider all options before making a prediction about the next event. 2. LSTM LSTM stands for Long short-term memory (LSTM). It is a methodology which works similar to that of neural network which is used for generating different types of tones, songs, etc based on the input given by the user. Architecture of LSTM: The diagram give below is the architectural view of the LSTM. It gives full overview about the working and the methodology of the LSTM. A typical implementation of an LSTM block. Steps Involved in LSTM Working: 1) Learning Chords: LSTM is used to learn musical chord structure. Chords structure of LSTM: LSTM was trained using a form of blues common in jazz bebop improvisation. The form is in 12 bars and contains the following chords. Chords having 12 bars 327 Copyright Vandana Publications. All Rights Reserved.

3 2) Learning Melody and chords: Initially chord structure must be learnt fully. The network is started with a series of notes and then notes are compose freely. Chords composed by LSTM Working of LSTM :- Structure of Feed Forward Neural Network 4. RECURRENT NEURAL NETWORKS The recurrent neural network is a type of neural network which works on the methodology of single input single output. Each output produced by the process becomes input to the other process. This is done recurrently to obtain a melodious song. 3. FEED FORWARD NEURAL NETWORKS Feed forward neural networks works on the concept of accepting the few strings from the user and these strings are processed using the probability distribution based algorithm. The new string generated is further processed using the current input and the other inputs by the user the process is fully based on the probability distribution algorithm. Consider an example, if the user inputs Dha Dhin Dha Dhin then the processing is done using the concept of the probability distribution algorithm. The output generated is not dependent on the past strings generated. This generated output is combination of the input strings. But only those values are displayed that satisfy the rules of the feed forward neural networks. Structure of Recurring Neural Network 328 Copyright Vandana Publications. All Rights Reserved.

4 The multiple feed forward neural networks are used to generate the output for the complex values. The string generated from above example is Dha Dhin Dhin Dha. This is used in generation of songs, tones, lyrics, etc. V. EXAMPLE Consider an example of certain tabla bols to demonstrate that how the system will generate its corresponding strings. IV. Structure of Neural Networks SYSTEM BLOCK DIAGRAM Dha Dhin Dhin Dha Dha Dhin Dhin Dha Dha Tin Tin Ta Ta Dhin Dhin Dha Teen Taal has 16 Bits and it has 4 sections with 4 beats in each section we can fit a tihai as follows in 16 Beats of Teen-Taal. DhaDha TeTe DhaDha TinNa 1 Dha # DhaDha TeTe DhaDha TinNa Dha # DhaDha TeTe DhaDha TinNa 16 Dha First Bit of time Cycle of teen Taal # represents pause As shown above N-gram modeling is applied to define bi-grams & tri-grams if required to predict the next alphabet which should have the best possible vocab as well as enough rhythmic. This project serves as an assistant for music artists & tabla masters and glitters our Indian culture over western methologies. V. CONCLUSION The above block diagram gives us the in general overview regarding our system and helps us to analyze how different modules are interred related with each other. The different modules of our system are Generator, Modifier, Authenticator, Player, etc. The main intention of this project is to develop a system which uses a derivation of an existing formal grammar(s) and to use it to model the language of Tabla compositions. The system will generate a set of compositions based upon some preliminary parameters and to be able to generate the same compositions in any taal. In order to make this a usable system, a front end is to be implemented whereby the user could select the parameters and the composition that is generated would be displayed (as text or audio samples) back to the user. This would serve the purpose of being an intelligent system as well as serving a practical purpose for the listener. The system which will be designed in this project will provide musical notes to the user. It will generate required mathematical rule base (theka or taal) using artificial intelligence and neural network techniques. The strings which will be generated can be melodious. Artist will be able to play & enhance the generated strings. 329 Copyright Vandana Publications. All Rights Reserved.

5 VI. ACKNOWLEDGEMENT We are thankful that our project mentor Mr. Prashant Kanade gave us opportunity to design such a system which recognizes grammar as well as follow the literature rules of tabla instrument and generate its respective combination of strings also called as Taal. As of now, this idea has not been explored yet in terms of musical notes generation specifically for tabla. So our idea will act as sowing the seed of automation in terms of rhythmic notes formation which help typical tabla masters to explore their knowledge. This project serves as an assistant for music artists & tabla masters and glitters our Indian culture over western methologies. REFERENCES [1] Recurrent LSTM Neural Networks for Language Modeling, [2] Wonmin Byeon, Thomas, M. Breuel, Federico Raue, Marcus Liwicki, Scene Labeling with LSTM Recurrent Neural Networks, 2015 IEEE Explore. [3] Benard Bel., Pattern grammars in formal representations of musical structures. In Work- shop on AI and Music, pages th International Joint Conference on Artificial Intelligence, June [4] Rowe, R, Interactive Music Systems - Machine Listening and Composing, The MIT Press, Mas, p. 26 [5] Cheng Long, Raymond Chi-Wing Wong, Raymond Ka Wai Sze, T-Music: A Melody Composer based on Frequent Pattern Mining, ICDE Conference 2013 [6] Chuan-Wang Chang and Hewijin Christine Jiau, An Improved Music Representation Method by Using Harmonic-based Chord Decision Algorithm, 2004 IEEE International Conference on Multimedia and Expo (ICME) [7] Abhijit Suprem and Manjit Ruprem, A New Composition Algorithm for Automatic Generation of Thematic Music from the Existing Music Pieces, Proceedings of the World Congress on Engineering and Computer Science 2013 Vol II WCECS 2013, October, 2013, San Francisco, USA [8] Spangler, R. R.,1999, Rule based Analysis and Generation of music, Ph.D., thesis, California Institute of Technology, California [9] Laura Lynn Housley, 2012, Dynamic Generation of Musical Notation from MusicXML Input on an Android Tablet, thesis, The Ohio State University [10] Miguel Haruki Yamaguchi, May 2011, An extensible tool for automated music generation, thesis, Lafayette College. S.Jorda, O. Wust, A System for Collaborative Music Composition over the Web, IEEE Explore 330 Copyright Vandana Publications. All Rights Reserved.