A euclidean distance between an input pattern and a reference pattern is elementwise determined, to sequentially calculate a set of cumulative values thereof, which is stored in a work memory for use in a current recurrence calculation and in a cumulative data memory for use in a decision for a speech recognition. The recurrence calculation is executed on one of division blocks obtained by collecting input frames and reference frames. Stored data in the work memory are shifted every block for a subsequent recurrence calculation.

Speech recognition system and speech recognition method with reduced response time for recognition

PURPOSE:To maintain the consistency of definition information so as to efficiently execute customizing at the time of customizing definition information on the frame of the work of software development. CONSTITUTION:A relation between items being the objects of customizing is defined as a related model. A support method for maintaining the consistency of definition information is allocated to the relation of the respective related models as a rule and it is stored in a related model storage part 5. The concrete definition of the rule is stored in a rule storage part 7. A customizing execution part 11 executes part 8. A customizing guiding part 10 learns an altered part from customizing history, refers to the related model and rule definition, automatically alters definition information on a related destination and indicates the alters part to a user or gives an alarm.

Customizing support system in software development

PROBLEM TO BE SOLVED: To provide a system which enables a user to have a conversation with a TV or other multimedia devices through a integrated access system where voice can be used. SOLUTION: A voice input from a user is evaluated by a speaker collation/ discrimination module, and parameters are read out from a user profile database on the basis of this evaluation. These parameters are adapted to a voice model of a voice recognizer, and a user-oriented conversation grammar is supplied to a natural language parser. The voice of the user is translated by the voice recognizer and the natural language parser for the purpose of discriminating the meaning of the voice input from the user to control a TV tuner. The parser is operated in cooperation with a command module, which is a medium for conversations with the user, and gives advice on a screen or a question by a synthesized voice in order to derive further input from the user if necessary.

Control system for audio/video device and integrated access system for control of audio/video constitution

PROBLEM TO BE SOLVED: To perform high speed Viterbi route calculation using HMM.SOLUTION: The optimum route searching device stores a transition likelihood γ supposing the state transition likelihood from a state i to a state j is equal and a transition likelihood β(>γ) which is a state transition likelihood from a transition source state i to a transition destination state j selected based on each transition destination state j. A partial likelihood state row estimation section includes: means to obtain an output likelihood that an event xis generated from a state based on the event xas a t-th observation; means that stores the likelihood Lin a t-1-th state i; means that calculates a product of the t-1-th maximum likelihood maxLand the transition likelihood γ as a first candidate; means that calculates a maximum value of the product of each likelihood Lof the t-1-th selected transition source state i and a transition likelihood β as a second candidate; and means that calculate the likelihood Lof the t-th state j from the product of a larger value in the first candidate and the second candidate and the output likelihood, obtains a maximum likelihood maxLfrom the likelihood Lin the state j to determine a state j having a maximum likelihood.

Optimum route searching device

The invention provides a speech recognition method and device, a piece of electronic equipment and a readable storage medium. The method includes the following steps: acquiring multiple speech framesobtained by dividing a speech file corresponding to a target person according to a preset frame length; generating a feature vector of each speech frame; clustering the feature vectors of the speech frames to generate feature center vectors; and determining feature center vectors containing speech information from the feature center vectors of the speech frames according to a feature center vectorof a reference noise frame, and generating a target speech feature center vector, wherein the reference noise frame is a noise frame in the multiple speech frames, and the target speech feature center vector is used to determine the identity information of the target person. The problem of slow recognition caused by de-noising and the problem of low accuracy of speech recognition caused by the inability to remove all noise are solved. Features can be extracted directly, and noise can be weakened. The speed and accuracy of recognition are improved.

Speech recognition method and device, electronic equipment and readable storage medium

Speaker recognition equipment

Continuous voice recognition equipment

PURPOSE:To cover dispersion of each utterance and heighten recognition rate by providing a memory section that stores a learning pattern, updating the learning pattern by a new learning pattern and preparing a reference pattern from the learning pattern CONSTITUTION:A recognition decision section 5 finds minimum value of degree of similarity between each reference pattern sent from a similarity calculating section 4 and an input pattern, and outputs a category in which the minimum value is obtained as the result of recognition A confirmation inputting section 6 outputs the result of recognition to an FIFO memory section 7 as a control signal when a user made confirmation decision A reference pattern preparing section 8 reads out learning sample for each category from the FIFO memory section 7 and finds the average pattern The average pattern obtained for each category is sent to a reference pattern memory 3 and the reference pattern is renewed

Watari Masao

Papers

Speaker recognition equipment

Continuous voice recognition equipment

Pattern recognizing device

Normalized cepstrum analyzer

Speaker checking system