Hitachi, Ltd.

Nippon Telegraph and Telephone Corporation

Manabu Okamoto

Takahiko Kozaki

Morihito Miyagi

Yoshito Sakurai

Kenichi Ohtsuki

Takaaki Fukutomi

Kiyoaki Matsui

Shigeaki Aoki

Yusuke Shinohara

Takashi Nakamura

Yoshikazu Yamaguchi

Yutaka Nishino

Hiroyuki Matsui

Takafumi Moriya

Ikuichiro Kinoshita

Mariko Aoki

Nobuo Hayashi

Provided is a technology of learning an acoustic model with a certain degree of accuracy of sound recognition within a short calculation period. An acoustic model learning device includes: a loss calculation unit configured to calculate a loss of sound data which is an element of the corpus Cfor learning by using an acoustic model; a curriculum corpus generation unit configured to generate a curriculum corpus being a union of subsets of the corpuses Cfor learning, the corpuses Cincluding, as elements, sound data for which the loss falls within a predetermined range indicating a small value; an acoustic model update unit configured to update the acoustic model by using the curriculum corpus; and a first end condition determination unit configured to output the acoustic model when a predetermined end condition is satisfied, or transfer execution control to the loss calculation unit when the predetermined end condition is not satisfied, and the acoustic model update unit is configured to update the acoustic model by giving a weight to a gradient for sound data which is an element of the curriculum corpus using such a weight for sound data as to have a smaller value as a number of times the sound data has been selected as an element of the curriculum corpus becomes larger.

Acoustic model learning apparatus, acoustic model learning method, and program

A training speech data generating apparatus includes: a voice conversion unit that converts, using fourth noise data, which is noise data based on third noise data, and speech data, the speech data so as to make the speech data clearly audible under a noise environment corresponding to the fourth noise data; and a noise superimposition unit that obtains training speech data by superimposing the third noise data and the converted speech data.

Learning speech data generating apparatus, learning speech data generating method, and program

An ATM switching system includes PVC allocation circuits-i corresponding to output queues-i is disclosed. At the time of arrival of the leading cell of each burst data, if the output line has a room, the PVC of the burst data is stored as information for admission and identification. The subsequently-arriving cells having the same PVC identification information as the stored identification information are 'input to an output buffer, and the cells of the remaining burst data are all discarded. Even when a plurality of burst data compete for the same output line, the fact that all the cells other than the first-arriving admitted burst data are discarded prevents the cells from being input other than within a predetermined bandwidth and localizes the burst data affected by the cell discarding due to output queue overflow.

ATM switching system and cell control method

An ATM switching system which includes PVC allocation circuits corresponding to output queues. At the time of arrival of the leading cell of each burst data, if the output line has room, the PVC of the burst data is stored as information for admission and identification. The subsequently-arriving cells, having the same PVC identification information as the stored identification information are input to an output buffer, and the cells of the remaining burst data are all discarded. Even when a plurality of burst data compete for the same output line, the fact that all the cells other than the first-arriving admitted burst data are discarded prevents the cells from being input other than within a predetermined bandwidth and localizes the burst data affected by the cell discarding due to output queue overflow.

An ATM switching system which includes PVC allocation circuits corresponding to output queues. At the time of arrival of the leading cell of each burst data, if the output line has room, the PVC of the burst data is stored as information for admission and identification. The subsequently-arriving cells having the same PVC identification information as the stored identification information are input to an output buffer, and the cells of the remaining burst data are all discarded. Even when a plurality of burst data compete for the same output line, the fact that all the cells other than the first-arriving admitted burst data are discarded prevents the cells from being input other than within a predetermined bandwidth and localizes the burst data affected by the cell discarding due to output queue overflow.

A time difference .DELTA..tau. between the arrival of acoustic signals from sound sources to microphones 1, 2 is detected from output channel signals L, R from microphones 1, 2. By Fourier transform, the signals L, R are divided into respective frequency bands L(f1)-L(fn), R(f1)-R(fn). Differences .DELTA..tau.i (i=1, 2, . . . n) in the time-of-arrival of L(f1)-L(fn) and R(f1)-R(fn) to the microphones 1, 2 as well as a signal level difference .DELTA.Li are detected. L(f1)-L(fn), R(f1)-R(fn) are divided into a low range of fi&lt;1/(2 .DELTA..tau.), a middle range of 1/(2 .DELTA..tau.) &lt;fi&lt;1/.DELTA..tau., and a high range of fi&gt;1/.DELTA..tau.. Utilizing .DELTA..tau.i for the low range, .DELTA.Li and .DELTA..tau.i for the middle range and .DELTA.Li for the high range, a determination is made from which sound source L(fi), R(fi) are oncoming to deliver outputs separately for each sound source. The outputs are subject to an inverse Fourier transform for synthesis separately for each sound source.

Method and apparatus for separation of source, program recorded medium

An ATM switching system which includes PVC allocation circuits corresponding to output queues is disclosed. At the time of arrival of the leading cell of each burst data, if the output line has room, the PVC of the burst data is stored as information for admission and identification. The subsequently-arriving cells having the same PVC identification information as the stored identification information are input to an output buffer, and the cells of the remaining burst data are all discarded. Even when a plurality of burst data compete for the same output line, the fact that all the cells other than the first-arriving admitted burst data are discarded prevents the cells from being input other than within a predetermined bandwidth and localizes the burst data affected by the cell discarding due to output queue overflow.

In an audio communication control unit, audio signals from at least three terminals connected to a switching part are branched to left- and right-channels audio signals, then the left- and right-channel audio signals are processed using sound image control parameters in such a manner as to impart them with target positions different for each terminal, then all left-channel audio signals corresponding to the respective terminals are mixed together into a left-channel mixed audio signal and all the right-channel audio signals are mixed together into a right-channel mixed audio signal, and these left- and right-channel mixed audio signals are distributed and sent to all the connected terminals.

Audio communication control unit

An optical switching system to which a plurality of input transmission lines and output transmission lines for transmitting optical signals are arranged and connected, and in which fixed length cells transmitted from the input transmission lines are switched based on the routing information contained in the cells and the switched cells are outputted to the output transmission lines includes: optical cell synchronizer portions for detecting the fronts of the cells and synchronizing the fronts of the cells incoming from respective transmission lines; a switching portion for converting respective cells the fronts of which are synchronized to those of different wavelengths by input transmission lines and for switching them based on the routing information in the cells; multiplexer portions for wavelength division multiplexing the cells switched to different wavelengths by input transmission line with respect to respective output transmission lines; demultiplexer portions for splitting cells wavelength division multiplexed in the multiplexer portions by cell and aligning them in a time series, and wavelength converter portions for converting the cells split in the demultiplexer portions to cells having a predetermined single output wavelength.

Growing community of inventors

Yokohama, Japan

Manabu Okamoto