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.
Type:
Grant
Filed:
March 3, 1992
Date of Patent:
September 10, 1996
Assignee:
ATR Interpreting Telephony Research Laboratories
Abstract: A system and method are provided for improving quality of sound generated by computerized systems having limited memory. A wavetable synthesizer is implemented wherein data utilized to synthetically generate acoustic waveforms is stored. A plurality of datasets is also generated and stored, each comprised of a digitized acoustic waveform. In response to a MIDI datastream, the system determines if an appropriate stored acoustic sample corresponding thereto resides in the system's memory. If so, the system will generate the desired sound utilizing the stored acoustic sample data. If not, the system automatically determines in real time the appropriate wavetable dataset which will generate a sound most closely approximating the acoustic sound. The system thus dynamically reconfigures in real time between wavetable and acoustic sample synthesis, being configured for the former when appropriate acoustic samples are not present.
Type:
Grant
Filed:
December 3, 1992
Date of Patent:
August 22, 1995
Assignee:
International Business Machines Corporation
Abstract: A system for adaptively processing a telephonic speech signal performs modification in either the spectral domain or the time domain to bring the power in each frequency above the hearing threshold of the listener but below the upper limit of the listener's dynamic range.