Special Indices for Laa Lyric Analysis Generation Framework

Special Indices for Laa Lyric Analysis & Generation Framework Dr. V. Madhan Karky, Tamil Computing Lab (Ta. Co. La), College of Engineering Guindy, Anna University, Chennai.

Overview • • • Objective Introduction Background Rhyme Schemes in Tamil Meter Pattern System Architecture Indexing Structure Indexing Algorithm Results and Analysis

Objective • To build special indices for the Laa lyric analysis and generation framework to facilitate faster retrieval based on – Meter Pattern – Rhyme

Introduction • Tamil is a vibrant language with a rich grammar, vocabulary, an inherent poetic flavor. • About 1000 lyrics are being written every year as private albums, jingles and as original soundtracks of mainstream movies.

Background • WASP (Pablo Grevas (2000)) splits a given block of text, identifies patterns and fits words from the vocabulary to get verses of similar pattern. • COLIBRI (Agudo, Grevas, Calero (2002)) follows a case based approach to poem generation.

Background • Tra-la-Lyrics (Oliviera, Cardoso, Pereira (2005)) finds out the beat pattern of the midi file and places words with similar syllabic division and stress pattern. • Automatic generation of tamil lyrics for melodies (Ramakrishnan, Kuppan, Devi (2009)) converts the midi file to KNM reprsn and fits words to it from a corpus. Phrases are meaningful as parts only. No edhugai, monai or iyaybu.

Background • Laa (Sowmiya, Karky (2010)) talks of splitting raw text from midi file to templates and filling them with words from a wordnet according to the pattern mined from an existing corpus of lyrics with due consideration to rhyme, meaning and flow.

Rhyme Schemes in Tamil • Two words are said to rhyme in – Monai (���� ) - first letters are the same. – Edhugai (����� ) - second letters are the same. – Iyaibu (����� ) - last letters are the same.

Rhyme Schemes in Tamil • Examples: and ����� rhyme in monai. – ����� and ����rhyme in edhugai. – ������ and ���� rhyme in iyaibu. – ����� and ������ rhyme in edhugai and iyaibu. – ���� and ����� rhyme in all the three schemes. – ����

Meter Pattern • Maathirai (���� ) - time taken to wink an eyelid. • Maathirai based classification of Tamil alphabets. – Nedil (N) (������) - Those alphabets which are pronounced for the time interval of 2 maathirai. – Kuril (K) (������) - Alphabets which take 1 maathirai to be pronounced. – Mei (M) (���� ) - Alphabets which are pronounced for 0. 5 maathirai. • Meter pattern of a word refers to its Kuril Nedil Mei pattern. • For example, the Meter pattern of the word ����� is NKM as �� is a Nedil(N), � is a Kuril(K) and �� is a Mei(M).

System Architecture Word Object Convertor Lyric DB Rhyme Extractor Rhyme Pattern Extractor Index Builder Rhyme Meter Index

Indexing Structure Part of Speech Letter 1 Words Letter 2 Words Letter 1 Words Letter 2 Words Meter. Pattern 1 ���� Meter. Pattern 2 Meter. Pattern 1 ����� Meter. Pattern 2

Indexing Algorithm

Result and Analysis 4000 3500 3000 Meter Rhyme Indexed Approach Word Indexed Approach 2500 Retrieval Time (in milliseconds)2000 1500 1000 500 484 463 442 421 400 379 358 337 316 295 274 253 232 211 190 169 148 127 85 106 64 43 22 1 0 Word

Results and Analysis • Retrieval complexity of both the approaches tested using a dataset of 500 tamil words. • The average retrieval time in – Word indexed approach - 875. 47 millisecond – Meter Rhyme Indexed approach – 1. 90 millisecond

Results and Analysis • The drastic decrease in retrieval time from O(α) to O(1) [α is the number of words in the database] is due to – The use of hash-tables which are efficient for retrieval. – Having separate hash-tables for the ����� and ����� of each POS. ,

Thank You!!!