Download Towards an Objective Comparison of Panning Feature Algorithms for Unsupervised Learning Estimations of panning attributes are an important feature to extract from a piece of recorded music, with downstream uses such
as classification, quality assessment, and listening enhancement.
While several algorithms exist in the literature, there is currently
no comparison between them and no studies to suggest which one
is most suitable for any particular task. This paper compares four
algorithms for extracting amplitude panning features with respect
to their suitability for unsupervised learning. It finds synchronicities between them and analyses their results on a small set of
commercial music excerpts chosen for their distinct panning features. The ability of each algorithm to differentiate between the
tracks is analysed. The results can be used in future work to either
select the most appropriate panning feature algorithm or create a
version customized for a particular task.
Download A Comparison of Extended Source-Filter Models for Musical Signal Reconstruction Recently, we have witnessed an increasing use of the sourcefilter model in music analysis, which is achieved by integrating the source filter model into a non-negative matrix factorisation (NMF) framework or statistical models. The combination of the source-filter model and NMF framework reduces the number of free parameters needed and makes the model more flexible to extend. This paper compares four extended source-filter models: the source-filter-decay (SFD) model, the NMF with timefrequency activations (NMF-ARMA) model, the multi-excitation (ME) model and the source-filter model based on β-divergence (SFbeta model). The first two models represent the time-varying spectra by adding a loss filter and a time-varying filter, respectively. The latter two are extended by using multiple excitations and including a scale factor, respectively. The models are tested using sounds of 15 instruments from the RWC Music Database. Performance is evaluated based on the relative reconstruction error. The results show that the NMF-ARMA model outperforms other models, but uses the largest set of parameters.
Download Polyphonic Pitch Detection by Iterative Analysis of the Autocorrelation Function In this paper, a polyphonic pitch detection approach is presented, which is based on the iterative analysis of the autocorrelation function. The idea of a two-channel front-end with periodicity estimation by using the autocorrelation is inspired by an algorithm from Tolonen and Karjalainen. However, the analysis of the periodicity in the summary autocorrelation function is enhanced with a more advanced iterative peak picking and pruning procedure. The proposed algorithm is compared to other systems in an evaluation with common data sets and yields good results in the range of state of the art systems.
Download Latent Force Models for Sound: Learning Modal Synthesis Parameters and Excitation Functions from Audio Recordings Latent force models are a Bayesian learning technique that combine physical knowledge with dimensionality reduction — sets of coupled differential equations are modelled via shared dependence on a low-dimensional latent space. Analogously, modal sound synthesis is a technique that links physical knowledge about the vibration of objects to acoustic phenomena that can be observed in data. We apply latent force modelling to sinusoidal models of audio recordings, simultaneously inferring modal synthesis parameters (stiffness and damping) and the excitation or contact force required to reproduce the behaviour of the observed vibrational modes. Exposing this latent excitation function to the user constitutes a controllable synthesis method that runs in real time and enables sound morphing through interpolation of learnt parameters.
Download A Corpus-Driven Parametric Modal Reverberator A parametric modal reverberator is presented in which synthesis parameters are derived from a large, curated corpus of room impulse responses (IRs). The collected responses are subjected to modal decomposition, yielding per-mode frequencies, damping coefficients, and residue amplitudes, together with a short early-reflection finite impulse response (FIR) filter. From the decomposed data, a feature table is constructed per IR comprising standard acoustic indices, per-band damping and density statistics, amplitude distributions, and FIR descriptors—50 variables in total. Six acoustically meaningful user controls are selected; since these exhibit substantial pairwise correlations across the corpus, they are orthogonalised via principal component analysis (PCA) prior to regression.
Download Characterisation of Acoustic Scenes Using a Temporally-constrained Shift-invariant Model In this paper, we propose a method for modeling and classifying acoustic scenes using temporally-constrained shift-invariant probabilistic latent component analysis (SIPLCA). SIPLCA can be used for extracting time-frequency patches from spectrograms in an unsupervised manner. Component-wise hidden Markov models are incorporated to the SIPLCA formulation for enforcing temporal constraints on the activation of each acoustic component. The time-frequency patches are converted to cepstral coefficients in order to provide a compact representation of acoustic events within a scene. Experiments are made using a corpus of train station recordings, classified into 6 scene classes. Results show that the proposed model is able to model salient events within a scene and outperforms the non-negative matrix factorization algorithm for the same task. In addition, it is demonstrated that the use of temporal constraints can lead to improved performance.
Download The Tonalness Spectrum: Feature-Based Estimation of Tonal Components The tonalness spectrum shows the likelihood of a spectral bin being part of a tonal or non-tonal component. It is a non-binary measure based on a set of established spectral features. An easily extensible framework for the computation, selection, and combination of features is introduced. The results are evaluated and compared in two ways. First with a data set of synthetically generated signals but also with real music signals in the context of a typical MIR application.
Download Unsupervised Audio Key and Chord Recognition This paper presents a new methodology for determining chords of a music piece without using training data. Specifically, we introduce: 1) a wavelet-based audio denoising component to enhance a chroma-based feature extraction framework, 2) an unsupervised key recognition component to extract a bag of local keys, 3) a chord recognizer using estimated local keys to adjust the chromagram based on a set of well-known tonal profiles to recognize chords on a frame-by-frame basis. We aim to recognize 5 classes of chords (major, minor, diminished, augmented, suspended) and 1 N (no chord or silence). We demonstrate the performance of the proposed approach using 175 Beatles’ songs which we achieved 75% in F-measure for estimating a bag of local keys and at least 68.2% accuracy on chords without discarding any audio segments or the use of other musical elements. The experimental results also show that the wavelet-based denoiser improves the chord recognition rate by approximately 4% over that of other chroma features.
Download On the window-disjoint-orthogonality of speech sources in reverberant humanoid scenarios Many speech source separation approaches are based on the assumption of orthogonality of speech sources in the time-frequency domain. The target speech source is demixed from the mixture by applying the ideal binary mask to the mixture. The time-frequency orthogonality of speech sources is investigated in detail only for anechoic and artificially mixed speech mixtures. This paper evaluates how the orthogonality of speech sources decreases when using a realistic reverberant humanoid recording setup and indicates strategies to enhance the separation capabilities of algorithms based on ideal binary masks under these conditions. It is shown that the SIR of the target source demixed from the mixture using the ideal binary mask decreases by approximately 3 dB for reverberation times of T60 = 0.6 s opposed to the anechoic scenario. For humanoid setups, the spatial distribution of the sources and the choice of the correct ear channel introduces differences in the SIR of further 3 dB, which leads to specific strategies to choose the best channel for demixing.
Download Local Key estimation Based on Harmonic and Metric Structures In this paper, we present a method for estimating the local keys of an audio signal. We propose to address the problem of local key finding by investigating the possible combination and extension of different previous proposed global key estimation approaches. The specificity of our approach is that we introduce key dependency on the harmonic and the metric structures. In this work, we focus on the relationship between the chord progression and the local key progression in a piece of music. A contribution of our work is that we address the problem of finding a good analysis window length for local key estimation by introducing information related to the metric structure in our model. Key estimation is not performed on empirical-chosen segment length but on segments that are adapted to the analyzed piece and independent from the tempo. We evaluate and analyze our results on a new database composed of classical music pieces.