1,721,049 research outputs found
Recognition of musical instruments by statistical classification
No full textThe correct classification of single musical sources is a relevant aspect for the source separation task and the
automatic transcription of polyphonic music. In this paper, we present a classification experiment on six different musical
instruments: violin, clarinet, flute, oboe, saxophone and piano. It is characterized by two steps. In the first step, a suitable
signal preprocessing based on FFT and QFT (Q-constant Frequency Transform) is adopted for feature extraction and data set
preparation. In the second step, a nonexclusive classification method is proposed to handle the inevitable overlapping among
classes. It is obtained by a co-operative clustering technique. The success of this kind of classification method is conditioned
by the adopted clustering procedure. We propose a hierarchical scale-based approach for this task, carrying out good results
Recognition of musical chord notes
No full textWe developed a new algorithm for automatic recognition of musical chord notes. In the chord recognition field,
we often refer to the “Chord Spectrum”, that is the chord representation in the frequency domain. The main concept on which
the “chord recognition” theory is based is that we need to find similar groups of sinusoidal tones (tone patterns) belonging to
the chord spectrum, through which we can describe the chord as an acoustic profile, with the help of “generative subspectra”.
The work done by A. Tanguiane during last decade was the starting point to our study that considers chords played by one or
more instruments. In his research, Tanguiane described the chord features and used the information taken from the
autocorrelation of chord frequency components to recognize it. He considered that partials forming the chord was equally
spaced in a logarithmic way in the frequency domain, implying equal distances correspond to equal musical intervals.
To obtain the components equally spaced in a logarithmic way, we use the QFT (Q-constant Fourier Transform), introduced
by J.C. Brown in 1991, and also used for the recognition of the fundamental frequency of each note. The QFT allows to show
the energy of the singular frequencies in a logarithmic scale spectrum. Autocorrelation is evaluated over the components of
the QFT that exceed a certain threshold value.
The developed algorithm allows us to obtain good results both for recognition of two, three and four notes chords
Speech noise reduction using adaptive spline neural networks
No full textA new Neural Network architecture for real-time oriented speech denoising is proposed. It is based on Adaptive
Spline neurons, whose peculiarity is the adaptive activation function. So, in the training phase, we can update both values of
weights and activation function shape, obtaining networks with more flexibility and generalization capabilities. Net training is
performed through the classical back-propagation rule. We focused our attention to continuous uncorrelated disturbs and we
tried two kinds of approach: in the first one we processed the whole signal by a single network, while in the second one we
operated a frequency sub-bands decomposition and we processed every sub-channel separately in a parallel way. The first
approach is less heavy but the second one gives better results, due to the fact that, in practical application, background noise
is frequency dependent. Results show improvements of Signal to Noise Ratio (SNR) and better performances in comparison
with classical denoising neural networks
Adaptive room acoustic response simulation: a virtual 3D application
No full textIn this paper we propose a method to simulate a 3D acoustical environment in which sound sources are positioned
in a well defined side. The spatial position that human brain assigns to a sound source is influenced by two main elements:
the reverberation, which is related to many factors, including the distance of the source and the type of the environment, and
the differences between the sound signals that reach the listener’s ears, related to the sound source angulation with respect to
the listener’s head. All this elements have to be simulated in order to give the illusion that the instrument sound comes from a
particular position in a particular environment. To obtain this result, the proposed method requires to get the stereo Impulse
Response (IR) of the environment for each position that have to be simulated. Then, we approximate each IR by a
corresponding adaptive IIR filter, that performs real-time operations
Automatic recognition of piano music compositional styles
No full textIn this paper, we present a system for automatic recognition of piano music compositional styles. It is based on a classifier that takes as input a set of MIDI files containing music pieces composed by Bach, Mozart, Beethoven and Debussy. Several features are extracted from MIDI data: the number of times each degree of the tempered scale is repeated in the whole piece, the interval between a note and the preceding one, and finally a feature based on the Forti's representation of chords. Each piece of a training set is associated with the corresponding author. After a learning phase, the classifier is able to recognize the author of an unknown piece, by analyzing and classifying the feature set extracted from the unknown piece
Locally Connected BSB Neural Networks as Associative Memories Storing Grey-Scale Images
No full textIn this paper, we introduce an associative memory storing grey scale images. It’s based on a suitable translation of
the grey scale image into a Gray-coded binary image, stored in a single BSB binary neural network. The particular BSB we
are going to exploit has the property of local connectivity. The chosen learning algorithm guarantees asymptotic stability of
the stored patterns, low computational cost, and control of the connection weights precision, without multiplication
An improved method for CNN-based detection of symmetry axis in black and white images
No full textIn this paper, a method for symmetry axis detection in binary images is presented. The method is an improvement of a previous method presented by the same authors. The method exploits the nonlinear dynamic behavior of Cellular Neural Networks (CNNs), in particular the propagation of bipolar waves. The image is represented in polar form, transforming the symmetry with respect to an arbitrarily oriented axis in a vertical symmetry: the position of the vertical axis corresponds to the angle of the original symmetry axis. The parallel CNN architecture is useful to speed up the computation, because of the high computational cost of the task. The proposed algorithm is tested on many real images, with good results
Neural associative memory storing gray-coded gray-scale images
Full text linkIn this paper, we present a neural associative memory storing gray-scale images. The proposed approach is based on a suitable decomposition of the gray-scale image into, gray-coded binary images, stored in brain-state-in-a-box-type binary neural networks. Both learning and recall can be implemented by parallel computation, with time saving. The learning algorithm, used to store the binary images, guarantees asymptotic stability of the stored patterns, low computational cost, and control of the weights precision. Some design examples and computer simulations are presented to show the effectiveness of the proposed method
CNN based unsupervised pattern classification for linearly and non linearly separable data sets
No full textA novel algorithm for unsupervised classification of data sets made up of integer valued patterns by means of Cellular Neural Network (CNN) is proposed. The algorithm is suited both for linearly separable and non linearly separable data sets. The adopted CNN is n-dimensional and is based on a space-variant template - neighborhood order 1 - to cluster n-dimensional datasets. The choice of a CNN architecture allows a straightforward hardware implementation, particularly suited for bi-dimensional patterns
- …
