Beat tracking – i.e. deriving from a music audio signal a sequence of beat instants that might correspond to when a human listener would tap his foot – involves satisfying two con-straints: On the one hand, the selected instants should generally correspond to moments in the audio where a beat is indicated, for instance by the onset of a note played by one of the instru-ments. On the other hand, the set of beats should reflect a locally-constant inter-beat-interval, since it is this regular spacing between beat times that defines musical rhythm. These dual constraints map neatly onto the two constraints optimized in dynamic programming, the local match, and the transition cost. We describe a beat tracking system which first estimates a global tempo, uses this tempo to construct a transition cost function, then uses dynamic programming to find the best-scoring set of beat times that reflect the tempo as well as corresponding to moments of high ‘onset strength ’ in a function derived from the audio. This very simple and computationally efficient procedure is shown to perform well on the MIREX-06 beat track-ing training data, achieving an average beat accuracy of just under 60 % on the development data. We also examine the impact of the assumption of a fixed target tempo, and show that the system is typically able to track tempo changes in a range of ±10 % of the target tempo. 1 1

Abstract not found

This paper presents a system for recovering the sectional form of a musical piece: segmentation and labelling of musical parts such as chorus or verse. The system uses three types of acoustic features: mel-frequency cepstral coefficients, chroma, and rhythmogram. An analysed piece is first subdivided into a large amount of potential segments. The distance between each two segments is then calculated and the value is transformed to a probability that the two segments are occurrences of a same musical part. Different features are combined in the probability space and are used to define a fitness measure for a candidate structure description. Musicological knowledge of the temporal dependencies between the parts is integrated into the fitness measure. A novel search algorithm is presented for finding the description that maximises the fitness measure. The system is evaluated with a data set of 557 manually annotated popular music pieces. The results suggest that integrating the musicological model to the fitness measure leads to a more reliable labelling of the parts than performing the labelling as a post-processing step. 1

Abstract not found

We present a system to produce expectations based on the observation of a rhythmic music signals at a constant tempo. The algorithms we use are causal, in order be fit closer to cognitive constraints and allow a future realtime implementation. In a first step, an acoustic front-end based on the aubio library extracts onsets and beats from the incoming signal. The extracted onsets are then encoded in a symbolic way using an unsupervised scheme: each hit is assigned a timbre cluster based on its timbre features, while its inter-onset interval regarding the previous hit is computed as a proportion of the extracted tempo period and assigned an inter-onset interval cluster. In a later step, the representation of each hit is sent to an expectation module, which learns the statistics of the symbolic sequence. Hence, at each musical hit, the system produces both what and when expectations regarding the next musical hit. For evaluating our system, we consider a weighted average F-measure, that takes into account the uncertainty associated with the unsupervised encoding of the musical sequence. We then present a preliminary experiment involving generated musical material and propose a roadmap in the context of this novel application field. 1.

This paper presents AudioCycle, a prototype application for browsing through music loop libraries. AudioCycle provides the user with a graphical view where the audio extracts are visualized and organized according to their similarity in terms of musical properties, such as timbre, harmony, and rhythm. The user is able to navigate in this visual representation, and listen to individual audio extracts, searching for those of interest. AudioCycle draws from a range of technologies, including audio analysis from music information retrieval research, 3D visualization, spatial auditory rendering, audio time-scaling and pitch modification. The proposed approach extends on previously described music and audio browsers. Concepts developped here will be of interest to DJs, remixers, musicians, soundtrack composers, but also sound designers and foley artists. Possible extension to multimedia libraries are also suggested. 1.

Abstract not found

We will discuss the methodology and accuracy of the Echo Nest Musical Fingerprint (ENMFP), an open-source fingerprint code generator and query server that works on music files. We define fingerprinting as matching an arbitrary audio signal to its underlying song in a 10 million song database. 1. MUSIC FINGERPRINTING “Fingerprinting ” of audio files [1] [3] is becoming a necessary feature for any large scale music understanding service or system. Online music stores want to resolve existing user catalog against their cloud storage to save bandwidth. Music indexing tools want to adjust poor metadata. Music listeners want to remove duplicates and see necessary context next to their audio. The Echo Nest Corporation

The papers at this Convention have been selected on the basis of a submitted abstract and extended precis that have been peer reviewed by at least two qualified anonymous reviewers. This convention paper has been reproduced from the author’s advance manuscript, without editing, corrections, or consideration by the Review Board. The AES takes no responsibility for the contents. Additional papers may be obtained by sending request and remittance to Audio

Many interesting pieces of music violate established structures or rules of their genre on purpose. These songs can be very atypical in their interior structure and their different parts might actually allude to entirely different other songs or genres. We present a query-by-example-based user interface that shows songs related to the one currently playing. This relation is not based on overall similarity, but on the similarity between the part currently playing and parts of other songs in the collection along different dimensions (pitch, timbre, bars, beats, loudness). The similarity is initially computed automatically, but can be corrected by the user. Once a sufficient number of corrections has been made, we expect the similarity measure to reach an even higher precision. Our system thereby allows users to discover hidden similarities on the level of song sections instead of whole songs. 1.