Speech recognition system having an interface to a host computer bus for direct access to the host memory
First Claim
1. A speech recognition system for storing information on probability values of Markov models defining speech models of vocabulary words and information on a set of labels indicating phone types assignable to a micro period of time, for sequentially generating corresponding labels in response to an input speech to generate a label stream of said input speech, for accumulating said probability values of said Markov models on the basis of said speech models of said words and said label stream, and for performing speech recognition on the basis of the accumulated values, comprising:
- a bus;
a processor coupled to said bus;
a main memory for said processor for storing said probability values of said Markov models accessibly to each of said labels, said main memory being coupled to said bus; and
a signal processing unit coupled to said bus,said signal processing unit comprising;
means for storing information on said set of labels;
means for converting an input speech into said label stream on the basis of said information on the set of labels;
means for fetching said probability values of a Markov model related to each of labels contained in said label stream by direct memory access transfer from said main memory;
means for accumulating said probability values thus fetched for each of said vocabulary words; and
means for determining one or more words to be the result of the recognition on the basis of accumulated probability values.
2 Assignments
0 Petitions
Accused Products
Abstract
A signal processing card packaged on a bus of a personal computer has a bus master which is used to access the main memory of the personal computer. A large table of probability values required for speech recognition is held in the main memory. When a label to be processed is generated, only the necessary part of the table is read from the main memory to the memory on the signal processing card by direct memory access transfer to perform speech recognition processing.
255 Citations
9 Claims
-
1. A speech recognition system for storing information on probability values of Markov models defining speech models of vocabulary words and information on a set of labels indicating phone types assignable to a micro period of time, for sequentially generating corresponding labels in response to an input speech to generate a label stream of said input speech, for accumulating said probability values of said Markov models on the basis of said speech models of said words and said label stream, and for performing speech recognition on the basis of the accumulated values, comprising:
-
a bus; a processor coupled to said bus; a main memory for said processor for storing said probability values of said Markov models accessibly to each of said labels, said main memory being coupled to said bus; and a signal processing unit coupled to said bus, said signal processing unit comprising; means for storing information on said set of labels; means for converting an input speech into said label stream on the basis of said information on the set of labels; means for fetching said probability values of a Markov model related to each of labels contained in said label stream by direct memory access transfer from said main memory; means for accumulating said probability values thus fetched for each of said vocabulary words; and means for determining one or more words to be the result of the recognition on the basis of accumulated probability values. - View Dependent Claims (2, 3)
-
-
4. A speech recognition processing unit connectable to a bus of a computer system which computer system includes a processor, a main memory, and an input and output devices coupled in common to said bus, said unit comprising:
-
means for converting an input speech into a label stream on the basis of information on a set of labels indicating phone types assignable to a micro period of time; means for fetching probability values of Markov model related to each label contained in said label stream by direct memory access transfer from a main memory which beforehand stores information on said probability values of said Markov models defining speech models of vocabulary words, said information being arranged accessibly to each of said labels; means responsive to probability values thus fetched for accumulating said probability values for each of said words; and means responsive to the accumulated probability values for determining one or more words to be the result of the recognition.
-
-
5. A speech recognition system for storing information on a set of first labels indicating phone types assignable to a micro period of time, information on probabilities that respective ones of said first labels are produced in utterance of respective vocabulary words, information on speech models of said vocabulary words each represented as a chain of Markov models in a set which set includes Markov models corresponding to respective second labels indicating phone types assignable to a micro period of time, and information on probability values of said Markov models, in response to an input speech, for sequentially producing corresponding first labels to generate a first label stream of said input speech, for determining whether said vocabulary words are recognition candidates by using said first label streams and said probabilities that respective ones of said first labels are produced in utterance of respective vocabulary words, for accumulating said probability values of said Markov models on the basis of speech models of said recognition candidates and said first label stream, and for performing detailed speech recognition on the basis of accumulated probability values, comprising:
-
a bus; a processor coupled to said bus; a main memory for said processor for storing information of said probabilities that respective ones of said first labels are produced in utterance of respective ones of said vocabulary words, and strong information on speech models of each of said vocabulary words, said main memory being coupled to said bus; and a signal processing unit coupled to said bus, further said signal processing unit including; means for converting said input speech into said first label stream; means for fetching, for each of labels contained in said first label stream, information on probabilities that respective ones of said vocabulary words produce said first label, said information being fetched by direct memory access transfer from said main memory; means for accumulating probabilities thus fetched for said vocabulary word respectively upon each of said labels contained in said first label stream; means for determining candidate words on the basis of the accumulated probabilities; means, for each of first labels contained in said first label stream, for fetching information on values of probability that respective ones of said labels are produced at particular transitions in models of the candidate words, said information being fetched by direct memory access transfer from said main memory; means for computing likelihood that said word candidates be said input speech in response to said information on speech models and said information on probability values fetched for respective ones of said first labels; and means for generating a recognition result on the basis of said likelihood.
-
-
6. A speech recognition system comprising:
-
a general purpose computer system comprising a bus, a general purpose processor having a data signal connection to the bus, and a main memory having a data signal connection to the bus, said main memory storing label probability data for each of a plurality of vocabulary words; and an acoustic signal processor having a data signal connection to the bus of the general purpose computer system; characterized in that the acoustic signal processor comprises; means for measuring the value of at least one feature of an utterance over a series of successive time intervals to produce a series of label signals representing the feature values; a bus interface for accessing the main memory of the general purpose computer system through the bus without accessing the general purpose processor, said bus interface retrieving label probability data from the main memory; a match score processor for generating a match score for each vocabulary word from the label probability data for the vocabulary word obtained through the bus interface and from the series of label signals produced by the measuring means; and an output for outputting one or more vocabulary words having the best match scores. - View Dependent Claims (7, 8, 9)
-
Specification