Speech recognition circuit using parallel processors
A speech recognition circuit comprises an input buffer for receiving processed speech parameters. A lexical memory contains lexical data for word recognition. The lexical data comprises a plurality of lexical tree data structures. Each lexical tree data structure comprises a model of words having common prefix components. An initial component of each lexical tree structure is unique. A plurality of lexical tree processors are connected in parallel to the input buffer for processing the speech parameters in parallel to perform parallel lexical tree processing for word recognition by accessing the lexical data in the lexical memory. A results memory is connected to the lexical tree processors for storing processing results from the lexical tree processors and lexical tree identifiers to identify lexical trees to be processed by the lexical tree processors. A controller controls the lexical tree processors to process lexical trees identified in the results memory by performing parallel processing on a plurality of said lexical tree data structures.
1. An apparatus comprising:
a search controller, being a processor operating in dependence upon a program stored in a program memory to control a speech recognition search process; and
a plurality of lexical tree processors operating in parallel to perform the speech recognition search process, said lexical tree processors performing tasks initiated at the direction of the search controller;
wherein the search controller is configured to direct one or more of the lexical tree processors to perform initial estimates of scores for a speech parameter evaluated against a first set of lexical trees;
wherein the search controller is further configured to perform a pruning operation using the initial estimates of scores and to direct one or more of the lexical tree processors to perform a complete evaluation of scores evaluated against a second set of lexical trees;
wherein the first set of lexical trees is substantially broader than the second set of lexical trees; and
wherein the initial estimates of scores is performed substantially faster than the complete evaluation of scores.
2. An apparatus according to claim 1 , wherein the speech parameter is a feature vector.
3. An apparatus according to claim 1 wherein the speech parameter is a set of feature vectors derived from a set of audio samples representing sound captured over a finite time period.
4. An apparatus according to claim 1 , wherein the lexical tree processors receive from the search controller a pointer to lexical tree data or an identifier to lexical tree data.
5. A method of speech recognition using an apparatus having a search controller processor operating in dependence upon a program stored in a program memory, and a plurality of lexical tree processors operating in parallel to perform a speech recognition search process, said lexical tree processors performing tasks initiated at the direction of the search controller, the method comprising the steps of:
(i) performing initial estimates of scores for a speech parameter evaluated against a first set of lexical trees by one or more of the lexical tree processors directed by the search controller;
(ii) performing a pruning operation using the initial estimates of scores by the search controller; and
(iii) performing a complete evaluation of scores evaluated against a second set of lexical trees by the one or more of the lexical tree processors directed by the search controller;
wherein the first set of lexical trees is substantially broader than the second set of lexical trees; and
wherein the initial estimates of scores is performed substantially faster than the complete evaluation of scores.
6. A method according to claim 5 , wherein the speech parameter is a feature vector.
7. A method according to claim 5 , wherein the speech parameter is a set of feature vectors derived from a set of audio samples representing sound captured over a finite time period.
8. A method according to claim 5 , wherein during step (i) the search controller supplies to each said lexical tree processor a pointer to lexical tree data or an identifier to lexical tree data.
9. A method according to claim 8 , wherein the pointer or the identifier specifies the specific lexical tree node at which said lexical tree processor commences evaluation of state transitions.