Abstract: The automatic performance device includes a storage part for storing a plurality of performance patterns, a performance part for playing a performance on the basis of a performance pattern stored in the storage part, an input part for inputting performance information through an input device that receives a performance operation of a user, and a selection part for selecting a performance pattern being a maximum likelihood estimation from the plurality of performance patterns stored in the storage part on the basis of the performance information input to the input part.

Abstract: The automatic performance device includes a storage part for storing a plurality of performance patterns, a performance part for playing a performance on the basis of a performance pattern stored in the storage part, an input part for inputting performance information through an input device that receives a performance operation of a user, and a selection part for selecting a performance pattern being a maximum likelihood estimation from the plurality of performance patterns stored in the storage part on the basis of the performance information input to the input part.

Abstract: A method and apparatus multiplies a past sample a time lag ? older than a current sample by a quantized multiplier ?? on a frame by frame basis, subtracts the multiplication result from the current sample, codes the subtraction result, and codes the time lag using a fixed-length coder if the multiplier ?? is smaller than 0.2 or if information about the previous frame is unavailable, or codes the time lag using a variable-length coder if ?? is not smaller than 0.2. A multiplier ? is coded by a multiplier coder and the multiplier ?? obtained by decoding the multiplier ? is outputted. The process is performed for each frame.

Abstract: A method and apparatus multiplies a past sample a time lag ? older than a current sample by a quantized multiplier ?? on a frame by frame basis, subtracts the multiplication result from the current sample, codes the subtraction result, and codes the time lag using a fixed-length coder if the multiplier ?? is smaller than 0.2 or if information about the previous frame is unavailable, or codes the time lag using a variable-length coder if ?? is not smaller than 0.2. A multiplier ? is coded by a multiplier coder and the multiplier ?? obtained by decoding the multiplier ? is outputted. The process is performed for each frame.

Abstract: A method and apparatus multiplies a past sample a time lag ? older than a current sample by a quantized multiplier ?? on a frame by frame basis, subtracts the multiplication result from the current sample, codes the subtraction result, and codes the time lag using a fixed-length coder if the multiplier ?? is smaller than 0.2 or if information about the previous frame is unavailable, or codes the time lag using a variable-length coder if ?? is not smaller than 0.2. A multiplier ? is coded by a multiplier coder and the multiplier ?? obtained by decoding the multiplier ? is outputted. The process is performed for each frame.

Abstract: In difference coding, each of a first to M-th channel signals is divided into frames and independent energy of every channel signal and difference energy of difference signals between all channel signals are calculated for each frame. In ascending order of energy, if a signal corresponding to an energy value is independent signal, it is determined that independent coding should be used for the signal. If the signal is a difference signal and the type of coding for one of channel signals has been determined, it is determined that the other should be difference-coded using the former channel signal as a reference signal. If the type of coding for neither of the channel signals has been determined, it is determined that one of them should be independently coded and the other should be difference-coded using the former channel signal as a reference signal.

Abstract: The present invention obtains a separated signal from an audio signal based on the anisotropy of smoothness of spectral elements in the time-frequency domain. A spectrogram of the audio signal is assumed to be a sum of a plurality of sub-spectrograms, and smoothness of spectral elements of each sub-spectrogram in the time-frequency domain has directionality on the time-frequency plane. The method comprises obtaining a distribution coefficient for distributing spectral elements of said audio signal in the time-frequency domain to at least one sub-spectrogram based on the directionality of the smoothness of each sub-spectrogram on the time-frequency plane, and separating at least one sub-spectrogram from said spectral elements of said audio signal using said distribution coefficient.

Abstract: An object of the present invention is to efficiently perform weighted difference coding of two or more signals. Determination is made as to which of independent coding and weighted difference coding is to be used for each channel so that the total energy of the channel signals and weighted difference signals is minimized. A weighted difference signal is generated on the basis of the determination and a reference signal (parent) and a weight is generated as auxiliary codes, the difference signal is treated as an input channel signal, and the process of coding determination and difference signal and auxiliary code generation is repeated. The difference signal generated at the last iteration of the process and a signal to be coded by independent coding are compressive coded and the auxiliary codes generated at the iterations of the process are coded and outputted.

Abstract: An object of the present invention is to efficiently perform weighted difference coding of two or more signals. Determination is made as to which of independent coding and weighted difference coding is to be used for each channel so that the total energy of the channel signals and weighted difference signals is minimized. A weighted difference signal is generated on the basis of the determination and a reference signal (parent) and a weight is generated as auxiliary codes, the difference signal is treated as an input channel signal, and the process of coding determination and difference signal and auxiliary code generation is repeated. The difference signal generated at the last iteration of the process and a signal to be coded by independent coding are compressive coded and the auxiliary codes generated at the iterations of the process are coded and outputted.

Abstract: A method and apparatus multiplies a past sample a time lag ? older than a current sample by a quantized multiplier ?? on a frame by frame basis, subtracts the multiplication result from the current sample, codes the subtraction result, and codes the time lag using a fixed-length coder if the multiplier ?? is smaller than 0.2 or if information about the previous frame is unavailable, or codes the time lag using a variable-length coder if ?? is not smaller than 0.2. A multiplier ? is coded by a multiplier coder and the multiplier ?? obtained by decoding the multiplier ? is outputted. The process is performed for each frame.

Abstract: In difference coding, each of a first to M-th channel signals is divided into frames and independent energy of every channel signal and difference energy of difference signals between all channel signals are calculated for each frame. In ascending order of energy, if a signal corresponding to an energy value is independent signal, it is determined that independent coding should be used for the signal. If the signal is a difference signal and the type of coding for one of channel signals has been determined, it is determined that the other should be difference-coded using the former channel signal as a reference signal. If the type of coding for neither of the channel signals has been determined, it is determined that one of them should be independently coded and the other should be difference-coded using the former channel signal as a reference signal.

Abstract: A pattern recognition scheme using probabilistic models that are capable of reducing a calculation cost for the output probability while improving a recognition performance even when a number of mixture component distributions of respective states is small, by arranging distributions with low calculation cost and high expressive power as the mixture component distribution. In this pattern recognition scheme, a probability of each probabilistic model expressing features of each recognition category with respect to each input feature vector derived from each input signal is calculated, where the probabilistic model represents a feature parameter subspace in which feature vectors of each recognition category exist and the feature parameter subspace is expressed by using mixture distributions of one-dimensional discrete distributions with arbitrary distribution shapes which are arranged in respective dimensions.

Abstract: Continuous speech is recognized by selecting among hypotheses, consisting of candidates of symbol strings obtained by connecting phonemes corresponding to a Hidden Markov Model (HMM) having the highest probability, by referring to a phoneme context dependent type HMM from input speech using a HMM phoneme verification portion. A phoneme context dependent type LR (Left-Right) parser portion predicts a subsequent phoneme by referring to an action specifying item stored in an LR (Left to Right) parsing table to predict a phoneme context around the predicted phoneme using an action specifying item of the LR table.

Abstract: A model adaptation scheme in the pattern recognition, which is capable of realizing a fast, real time model adaptation and improving the recognition performance. This model adaptation scheme determines a change in a parameter expressing a condition of pattern recognition and probabilistic model training between an initial condition at a time of acquiring training data used in obtaining a model parameter of each probabilistic model and a current condition at a time of actual recognition. Then, the probabilistic models are adapted by obtaining a model parameter after a condition change by updating a model parameter before a condition change according to the determined change, when the initial condition and the current condition are mismatching. The adaptation processing uses a Taylor expansion expressing a change in the model parameter in terms of a change in the parameter expressing the condition.

Abstract: One-dimensional normal distributions in respective dimensions of a continuous multi-dimensional normal distribution of each state of HMMs representing speech units mean and variance values are tied among similar one-dimensional distributions. As a result, the total number of normal distributions for representing the model is reduced without degrading recognition performance.

Abstract: Training data is LPC analyzed to obtain a feature parameter vector sequence, which is subjected to Viterbi segmentation using reference phoneme models to separate phonemes. Each piece of phoneme data is used to estimate a mean vector of the corresponding reference phoneme model by a maximum a posteriori estimation method. The adapted phoneme model and the corresponding reference phoneme model are used to estimate a mean vector for an unadapted phoneme model through interpolation by a vector field smoothing method. Alternatively, the mean vector of the adapted phoneme model is further smoothed by the vector field smoothing method. By this, an adapted model is obtained which has, as its parameters, the mean vector obtained for each phoneme and other corresponding parameters.

Abstract: In method for detecting a speech period in a high-noise environment, the variation in the spectrum of an input signal per unit time is calculated over an analysis frame period, and when the frequency of spectrum variation falls in a predetermined range, the input signal of that frame is decided to be a speech signal.

Abstract: An automated method of generating a subword model for speech recognition dependent on phoneme context for processing speech information using a Hidden Markov Model in which static features of speech and dynamic features of speech are modeled as a chain of a plurality of output probability density distributions. The method comprising determining a phoneme context class which is a model unit allocated to each model, the number of states used for representing each model, relationship of sharing of states among a plurality of models, and output probability density distribution of each model, by repeating splitting of a small number of states, provided in an initial Hidden Markov Model, based on a prescribed criterion on a probabilistic model.

Abstract: The present invention is a learning method of a neural network for identifying N category using a data set consisted of N categories, in which one learning sample is extracted from a learning sample set in step SP1, and the distances between the sample and all the learning samples are obtained in step SP2. The closest n samples are obtained for each category in step SP3, and similarity for each category is obtained using the distances from the samples and a similarity conversion function f(d)=exp (-.alpha..multidot.d.sup.2). In step SP4, the similarity for each category is used as a target signal for the extracted learning sample, and it returns to an initial state until target signals for all the learning samples are determined. When target signals are determined for all the learning samples, in step SP5, the neural network is subjected to learning by the back-propagation using the learning samples and the obtained target signals.

Abstract: Input speech of a reference speaker, who wants to convert his/her voice quality, and speech of a target speaker are converted into a digital signal by an analog to digital (A/D) converter. The digital signal is then subjected to speech analysis by a linear predictive coding (LPC) analyzer. Speech data of the reference speaker is processed into speech segments by a speech segmentation unit. A speech segment correspondence unit makes a dynamic programming (DP) based correspondence between the obtained speech segments and training speech data of the target speaker, thereby making a speech segment correspondence table. A speaker individuality conversion is made on the basis of the speech segment correspondence table by a speech individuality conversion and synthesis unit.