NLP Classification on English and Chinese Pop Music

Abstract

Through vocal melody extraction and midi conversion, this study isolates the pitches and rhythm as a tuple of event onsets of a large corpora of Chinese and English popular music (and the rap genre specifically). With language model classification (through self-trained word embedding layer as input to LSTM), the study will train a model to classify a given textual representation of a segment of vocal melody in the hope of proving the existence of the implicit rules behind writing and its intricate relationships with language and culture.