Nippon Telegraph and Telephone Corporation

Yuma Koizumi

Shoichiro Saito

Noboru Harada

Shin Murata

Kazunori Kobayashi

Hiroaki Ito

Masahiro Yasuda

Yuta Kawachi

Hisashi Uematsu

Keisuke Kinoshita

Akira Nakagawa

Marc Delcroix

Shoko Araki

Hitoshi Ohmuro

Kenta Niwa

Sakurako Yazawa

Tsubasa Ochiai

Masataka Yamaguchi

Provided is an abnormality estimation device capable of appropriately determining normal data appearing less frequently as normal. The abnormality estimation device includes an estimation unit that estimates an anomaly degree of an acoustic signal, by using an abnormality estimation model that is optimized while using a set of normal sounds and is optimized so as to minimize a difference between an anomaly degree of a normal sound appearing more frequently and an anomaly degree of a normal sound appearing less frequently.

Abnormality estimation device, abnormality estimation method, and program

A signal processing device includes processing circuitry configured to receive an input of extraction target information indicating which audio class of an audio signal is to be extracted from a mixture audio signal constituted by a mixture of audio signals of a plurality of audio classes, and output a result of extracting the audio signal of the audio class indicated by the extraction target information from the mixture audio signal, with a neural network by using a feature value of the mixture audio signal and the extraction target information.

Signal processing device, signal processing method, signal processing program, learning device, learning method, and learning program

An anomaly detection technique which realizes high accuracy while reducing cost required for normal model learning is provided. An anomaly detection apparatus includes an anomaly degree estimating unit configured to estimate an anomaly degree indicating a degree of anomaly of anomaly detection target equipment from sound emitted from the anomaly detection target equipment (hereinafter, referred to as anomaly detection target sound) based on association between a first probability distribution indicating distribution of normal sound emitted from one or more pieces of equipment different from the anomaly detection target equipment and normal sound emitted from the anomaly detection target equipment (hereinafter, referred to as normal sound for adaptive learning).

Anomaly detection apparatus, probability distribution learning apparatus, autoencoder learning apparatus, data transformation apparatus, and program

A possible region of encoding results of anomalous samples is limited. An encoder storage unitstores an encoder for projecting an input feature value into a latent space in which the latent space is a closed manifold, a normal distribution obtained by learning normal data and an anomalous distribution obtained by learning anomalous data are held on the manifold, and a decoder for reconstructing the output of the encoder. An encoding unitobtains a reconstruction result output by the decoder when a feature value of target data is input to the encoder. An anomaly score calculation unitcalculates an anomaly score of the target data based on distances between the reconstruction result and the normal distribution and distances between the reconstruction result and the anomalous distribution.

Anomaly detection apparatus, anomaly detection method, and program

The accuracy of unsupervised anomalous sound detection is improved. An anomalous sound generation unitgenerates a pseudo-anomalous sound by associating a sound obtained from a predetermined device with a probability distribution followed by a sound obtained from a desired device group. A threshold value setting unitsets a threshold value so that all anomaly scores calculated from the pseudo-anomalous sound are determined to be anomalous. A parameter update unituses an obtained normal sound and an obtained anomalous sound, or a threshold value and a value based on the obtained anomalous sound, to update a parameter so to reliably determine an anomalous sound as anomalous and minimize the probability of determining a normal sound as anomalous.

Sound generation apparatus, data generation apparatus, anomaly score calculation apparatus, and program

By appropriately registering a sound with an arbitrary length to be registered, erroneous determination is suppressed. A normal sound registration apparatusextracts a feature amount with a fixed-length from a time-series acoustic signal with a variable arbitrary length. A frequency conversion unitacquires a time-series frequency signal obtained by frequency-converting the time-series acoustic signal. A feature extraction unitextracts a feature amount from the time-series acoustic signal. The feature extraction unitis optimized to extract, from the time-series acoustic signal including at least a known normal sound and an unknown normal sound, a feature amount which reflects a feature of the unknown normal sound, and to extract, from the time-series acoustic signal including at least an anomalous sound and a normal sound, a feature amount which reflects a feature of the anomalous sound.

Feature extraction apparatus, anomaly score estimation apparatus, methods therefor, and program

A direction-of-arrival estimation device for achieving direction-of-arrival estimation which is robust against an SNR and in which an application range of a learning model is specific is provided. The device includes: a reverberation output unit configured to receive input of a real spectrogram extracted from a complex spectrogram of acoustic data and an acoustic intensity vector extracted from the complex spectrogram, and output an estimated reverberation component of the acoustic intensity vector; a noise suppression mask output unit configured to receive input of the real spectrogram and the acoustic intensity vector from which the reverberation component has been subtracted, and output a time frequency mask for noise suppression; and a sound source direction-of-arrival derivation unit configured to derive a sound source direction-of-arrival based on an acoustic intensity vector formed by applying the time frequency mask to the acoustic intensity vector from which the reverberation component has been subtracted.

Direction of arrival estimation apparatus, model learning apparatus, direction of arrival estimation method, model learning method, and program

An input of a first observation signal corresponding to an incoming signal from a first direction is received, an angular rotation operation of the first observation signal is performed to obtain a second observation signal corresponding to an incoming signal from a second direction that is different from the first direction and the second observation signal is added to a set of training data.

Training data extension apparatus, training data extension method, and program

Erroneous detection of erroneously determining a normal sound to be anomalous is suppressed. A registered normal sound detection apparatuscalculates anomaly score of an observed signal. A feature extraction unitextracts a feature amount with a fixed-length from a time-series acoustic signal with an arbitrary length. An anomaly score calculation unitcorrects the anomaly score calculated from the observed signal so that the higher similarity score between the observed signal and a registered normal sound is, the smaller a value of the anomaly score is. The anomaly score calculation unitcalculates the similarity score by a similarity score function learned by using the feature amount extracted from the time-series acoustic signal including at least a normal sound by the feature extraction unit

Anomaly score estimation apparatus, anomaly score estimation method, and program

Provided is an acoustic signal processing technique for performing a signal transformation suitable for desired signal processing (e.g., sound source enhancement processing) on a signal, and then performing the desired signal processing on the transformed signal. An acoustic signal processing device performs signal processing M which is a desired target on an input acoustic signal x. The acoustic signal processing device includes a transform unit that performs transform processing P on the acoustic signal x to obtain a first transform coefficient X; a signal processing unit that performs signal processing M corresponding to a desired target on the first transform coefficient X to obtain a second transform coefficient {circumflex over (&#x2003;)}S; and an inverse transform unit that performs inverse transform processing Pon the second transform coefficient {circumflex over (&#x2003;)}S to obtain an acoustic signal {circumflex over (&#x2003;)}s subjected to signal processing which is a desired target, wherein the transform processing P, the inverse transform processing P, and the signal processing M are simultaneously optimized.

Acoustic signal processing apparatus, method and program for the same

To provide an anomalous sound detection training technique by which a feature amount extraction function for detecting anomalous sound can be generated irrespective of whether training data for anomalous signals is available or not. An anomalous sound detection training apparatus includes: a first function updating unitthat updates a feature amount extraction function and an feature amount inverse transformation function, which are input, based on an optimization index of a variational autoencoder; an acoustic feature extraction unitthat extracts an acoustic feature of normal sound based on training data for normal sound; a normal sound model updating unitthat updates a normal sound model by using the acoustic feature that is extracted; a threshold updating unitthat obtains a threshold &#x3c6;corresponding to a false positive rate &#x3c1;, which has a predetermined value, by using the training data for normal sound and the feature amount extraction function that is input; and a second function updating unitthat updates the feature amount extraction function that is updated, based on a Neyman-Pearson-type optimization index defined by the threshold &#x3c6;that is obtained, and repeatedly performs processing of each of the above-mentioned units.

Anomalous sound detection apparatus, degree-of-anomaly calculation apparatus, anomalous sound generation apparatus, anomalous sound detection training apparatus, anomalous signal detection apparatus, anomalous signal detection training apparatus, and methods and programs therefor

An anomalous sound detection training apparatus includes: a first acoustic feature extraction unit that extracts an acoustic feature of normal sound based on training data for normal sound by using an acoustic feature extractor; a normal sound model updating unit that updates a normal sound model by using the acoustic feature extracted; a second acoustic feature extraction unit that extracts an acoustic feature of anomalous sound based on simulated anomalous sound and extracts the acoustic feature of normal sound based on the training data for normal sound by using the acoustic feature extractor; and an acoustic feature extractor updating unit that updates the acoustic feature extractor by using the acoustic feature of anomalous sound and the acoustic feature of normal sound that have been extracted, in which processing by the units is repeatedly performed.

Anomalous sound detection training apparatus, acoustic feature extraction apparatus, anomalous sound sampling apparatus, and methods and programs for the same

Accuracy of unsupervised anomalous sound detection is improved using a small number of pieces of anomalous sound data. A threshold deciding part () calculates an anomaly score for each of a plurality of pieces of anomalous sound data, using a normal model learned with normal sound data and an anomaly model expressing the pieces of anomalous sound data, and decides a minimum value among the anomaly scores as a threshold. A weight updating part () updates, using a plurality of pieces of normal sound data, the pieces of anomalous sound data and the threshold, weights of the anomaly model so that all the pieces of anomalous sound data are judged as anomalous, and probability of the pieces of normal sound data being judged as anomalous is minimized.

Anomalous sound detection apparatus, anomaly model learning apparatus, anomaly detection apparatus, anomalous sound detection method, anomalous sound generation apparatus, anomalous data generation apparatus, anomalous sound generation method and program

A noise estimation parameter learning device is provided according to which even in a large space causing a problem of the reverberation and the time frame difference, multiple microphones disposed at distant positions cooperate with each other, and a spectral subtraction method is executed, thereby allowing the target sound to be enhanced. A noise estimation parameter learning device for learning noise estimation parameters used to estimate noise included in observed signals through a plurality of microphones, the noise estimation parameter learning device comprising: a modeling part that models a probability distribution of observed signals of the predetermined microphone, models a probability distribution of time frame differences, and models a probability distribution of transfer function gains; a likelihood function setting part that sets a likelihood function pertaining to the time frame difference, and a likelihood function pertaining to the transfer function gain, based on the modeled probability distributions; and a parameter update part that alternately and repetitively updates two variables of two likelihood functions, and outputs the time frame difference and the transfer function gain that have converged, as the noise estimation parameters.

Target sound enhancement device, noise estimation parameter learning device, target sound enhancement method, noise estimation parameter learning method, and program

An acoustic signal is separated based on a difference in the distance from a sound source to a microphone. By using a filter obtained by associating a value corresponding to an estimated value of a short-distance acoustic signal which is obtained by using 'a predetermined function' from a second acoustic signal derived from signals collected by 'a plurality of microphones' and is emitted from a position close to &#x201c;the plurality of microphones&#x201d; with a value corresponding to an estimated value of a long-distance acoustic signal which is emitted from a position far from &#x201c;the plurality of microphones&#x201d;, a desired acoustic signal representing at least one of a sound emitted from a position close to &#x201c;a specific microphone&#x201d; and a sound emitted from a position far from &#x201c;the specific microphone&#x201d; is acquired from a first acoustic signal derived from a signal collected by &#x201c;the specific microphone&#x201d;. Note that &#x201c;the predetermined function&#x201d; is a function which uses such an approximation that a sound emitted from the position close to &#x201c;the plurality of microphones&#x201d; is collected as a spherical wave, and a sound emitted from the position far from &#x201c;the plurality of microphones&#x201d; is collected as a plane wave.

Growing community of inventors

Musashino, Japan

Yuma Koizumi