Automatic Transcription of Flamenco Singing From Polyphonic Music Recordings

Abstract

Automatic note-level transcription is considered one of the most challenging tasks in music information retrieval. The specific case of flamenco singing transcription poses a particular challenge due to its complex melodic progressions, intonation inaccuracies, the use of a high degree of ornamentation and the presence of guitar accompaniment. In this study, we explore the limitations of existing state of the art transcription systems for the case of flamenco singing and propose a specific solution for this genre: We first extract the predominant melody and apply a novel contour filtering process to eliminate segments of the pitch contour which originate from the guitar accompaniment. We formulate a set of onset detection functions based on volume and pitch characteristics to segment the resulting vocal pitch contour into discrete note events. A quantised pitch label is assigned to each note event by combining global pitch class probabilities with local pitch contour statistics. The proposed system outperforms state of the art singing transcription systems with respect to voicing accuracy, onset detection and overall performance when evaluated on flamenco singing datasets.

Corpus COFLA: A research corpus for the Computational study of Flamenco Music

“…The analysis window was set to N = 1024 samples with a hop size of h = 128 samples at a sampling rate of f s = 44.1kHz. We furthermore provide automatic note-level transcriptions of the singing voice melody obtained with the system described in [Kroher and Gómez 2015]. The transcriptions contain an onset time, duration and MIDI pitch value for each transcribed note and are provided as text and MIDI files.…”

“…Consequently, the authors report mistakenly transcribed guitar contours as a main source of error. Based on these findings, [Kroher and Gómez 2015] apply an additional contour filtering stage in order to eliminate contour sections which originate from the guitar accompaniment.…”

“…Related work in this relatively young field has mainly been carried out in the scope of the COFLA 1 research project. Apart from automatic transcription of the singing melody ([Gómez and Bonada 2013], [Kroher and Gómez 2015]), previous research has focused on melodic and rhythmic similarity ( [Kroher et al 2014], [Gómez-Martín et al 2015], [Díaz-B á ñez et al 2005]), melodic pattern detection ( [Pikrakis et al 2012] and ), metrical structure identification ( [Wu 2013]), perceived emotion ( [Tizón et al 2013]), genre classification ( ) and intra-and inter-style classification ([Díaz-B á ñez and Rizo 2014]). A comprehensive overview is provided in .…”

Musical note onset detection based on a spectral sparsity measure

“…As all these methods rely on differences between successive frames, the performance of these methods may drop when comparing successive notes sharing a considerable number of harmonics or with repeated notes having insufficient increase in magnitude at their onset. Finally, a more recent approach to NOD using pitch tracking was presented in [25] and performs better with pitched non-percussive instruments.…”

A Streamlined Encoder/decoder Architecture for Melody Extraction

“…Melody extraction is the task that aims to estimate the fundamental frequency (F0) of the dominant melody. Automatic melody extraction has been an active topic of research in the literature, since it has many important downstream applications in music analysis and retrieval [1][2][3][4].…”