BDB Bertsolaritzaren datu-basea

The availability of appropriate research corpora is a fundamental concern in music information retrieval research. This paper addresses the design, development, and evaluation of a poetic corpus, POMET, for the meter estimation task. Poems, which communicate through rhythm and apparent meaning, have a vital role in many literary traditions. Metrical rhythm generally involves periodic arrangements of sequences of stressed and unstressed syllables in each line of poems. It has already been proved that poetry’s aesthetic and emotional perception can be studied well using poetic meter analysis. A corpus with eight meters is designed and recorded in a studio environment for Malayalam, one of the prominent languages in South India. Using deep neural network architectures, a pilot evaluation is performed with musical texture features and spectrograms. We hope that the corpus can be used as a benchmark dataset for poetic meter estimation, rhythmic analysis, and corpus-based prosody analysis.