Music summarization using key phrases

Transcribing Debussy's Syrinx dynamics through Linguistic Description: The MUDELD algorithm

“…Summarization and description techniques have been used in the context of Music, although almost exclusively to the task of automatically providing a representative summary or 'key phrase' of a piece of music; that is, to find the most suitable and representative except that account for the whole piece [32]. This summarization, due to its particularities, is generally done by taking into account the low-level signal information from the audio.…”

Using duration models to reduce fragmentation in audio segmentation

“…Logan and Chu (2000) implicitly note, in the context of phrase summarization, one of these causes: a mismatch between the number of distinct segment labels requested and the information in the signal being segmented. If more labels are assigned than are required, then necessarily some of the desired segments will be mislabeled; in addition, each label will model a smaller volume of the feature space, which might cause individual segments to be fragmented to an undesired level of detail.…”

“…For instance, Foote proposed that segments should be defined as 'self-similar' intervals delimited by boundaries corresponding to peaks in a 'novelty' function computed by correlating a Gaussian-tapered 'checkerboard' kernel along the main diagonal of the dissimilarity matrix. Logan and Chu (2000), also using MFCCs as their spectral features, proposed a method for summarization employing both hidden-Markov models (HMMs) and threshold-based clustering methods, grouping features into key song segments. Peeters, Burthe, and Rodet (2002) propose a multi-pass clustering approach that uses both k-means and HMM-based clustering using multi-scale MFCC features.…”

Recent studies on music information processing

“…Chorus and phrase repetition, music structure On entering the 2000s, new approaches appeared based on the detection of similar sections that repeat within a musical piece (such as a repeating phrase). These led to methods for extracting the most representative section of a musical piece (usually the chorus) from one location [43][44][45]; music-summarization methods that shorten a musical piece leaving only main sections [46,47]; and the RefraiD method that exhaustively detects all chorus sections [48]. Among these methods, RefraiD focuses on chorus detection with the capability of determining the start and end points of every chorus section regardless of whether a key-change occurs.…”