# NoveltyCurve

standard mode | Rhythm category

## Inputs

frequencyBands(vector_vector_real) - the frequency bands

## Outputs

novelty(vector_real) - the novelty curve as a single vector

## Parameters

frameRate(real ∈ [1, ∞), default = 344.531) :- the sampling rate of the input audio

normalize(bool ∈ {true, false}, default = false) :- whether to normalize each band's energy

weightCurve(vector_real, default = []) :- vector containing the weights for each frequency band. Only if weightCurveType==supplied

weightCurveType(string ∈ {flat, triangle, inverse_triangle, parabola, inverse_parabola, linear, quadratic, inverse_quadratic, hybrid, supplied}, default = hybrid) :- the type of weighting to be used for the bands novelty

## Description

This algorithm computes the "novelty curve" (Grosche & Müller, 2009) onset detection function. The algorithm expects as an input a frame-wise sequence of frequency-bands energies or spectrum magnitudes as originally proposed in [1] (see FrequencyBands and Spectrum algorithms). Novelty in each band (or frequency bin) is computed as a derivative between log-compressed energy (magnitude) values in consequent frames. The overall novelty value is then computed as a weighted sum that can be configured using 'weightCurve' parameter. The resulting novelty curve can be used for beat tracking and onset detection (see BpmHistogram and Onsets).

Notes: - Recommended frame/hop size for spectrum computation is 2048/1024 samples (44.1 kHz sampling rate) [2]. - Log compression is applied with C=1000 as in [1]. - Frequency bands energies (see FrequencyBands) as well as bin magnitudes for the whole spectrum can be used as an input. The implementation for the original algorithm [2] works with spectrum bin magnitudes for which novelty functions are computed separately and are then summarized into bands. - In the case if 'weightCurve' is set to 'hybrid' a complex combination of flat, quadratic, linear and inverse quadratic weight curves is used. It was reported to improve performance of beat tracking in some informal in-house experiments (Note: this information is probably outdated).

- References:
- [1] P. Grosche and M. Müller, "A mid-level representation for capturing dominant tempo and pulse information in music recordings," in International Society for Music Information Retrieval Conference (ISMIR’09), 2009, pp. 189–194. [2] Tempogram Toolbox (Matlab implementation), http://resources.mpi-∞.mpg.de/MIR/tempogramtoolbox

## See also

BpmHistogram (standard) BpmHistogram (streaming) FrequencyBands (standard) FrequencyBands (streaming) Histogram (standard) Histogram (streaming) NoveltyCurve (streaming) Onsets (standard) Onsets (streaming) Spectrum (standard) Spectrum (streaming)