Towards Reference-Independent Rhythm Assessment of Solo Singing

Rhythm is an important aspect of singing in music information retrieval. From the principles of music theory, note duration is related to the time signature of a song, therefore it provides a representation for rhythm analysis. However, in a reference-independent setup, where the reference musical score of a song is not available, measurement of note duration and hence assessment of rhythm quality of solo singing is difficult. In this work, we analyse the correlation between the duration distributions of musical note lengths and singing syllable lengths, and propose a reference-independent framework that uses singing syllable duration distribution to assess the rhythm quality of solo singing. We show improved prediction of rhythm assessment score using syllable duration histogram compared to existing reference-independent frameworks.