Language model based on the speech recognition history
First Claim
1. A small vocabulary pattern recognition system for recognizing a sequence of words;
- the vocabulary storing a representation of a plurality of reference words;
the system comprising;
input means for receiving a time-sequential input pattern representative of a spoken or written word sequence;
a pattern recognizer comprising a word-level matching unit for generating a plurality of sequences of words by statistically comparing the input pattern to the representations of the reference words of the vocabulary;
a cache for storing a plurality of most recently recognized words and word sequence, said cache comprising a plurality of slots, wherein at least one or more slots are reserved for permanent use for storing infrequently retrieved user-designated critical word sequences that are spoken fully and not replaced by subsequently recognized words;
a sequence-level matching unit for selecting a word sequence from the plurality of sequences of words using a statistical language model which provides a probability of a word sequence;
the probability depending on a frequency of occurrence and non-occurrence of the word sequence in the cache.
2 Assignments
0 Petitions
Accused Products
Abstract
A small vocabulary pattern recognition system is used for recognizing a sequence of words, such as a sequence of digits (e.g. telephone number) or a sequence of commands. A representation of reference words is stored in a vocabulary 132, 134. Input means 110 are used for receiving a time-sequential input pattern representative of a spoken or written word sequence. A pattern recognizer 120 comprises a word-level matching unit 130 for generating a plurality of possible sequences of words by statistically comparing the input pattern to the representations of the reference words of the vocabulary 132, 134. A cache 150 is used for storing a plurality of most recently recognized words. A sequence-level matching unit 140 selects a word sequence from the plurality of sequences of words in dependence on a statistical language model which provides a probability of a sequence of M words, M≧2. The probability depends on a frequency of occurrence of the sequence in the cache. In this way for many small vocabulary systems where no reliable data is available on frequency of use of word sequences, the cache is used to provide data representative of the actual use.
-
Citations
9 Claims
-
1. A small vocabulary pattern recognition system for recognizing a sequence of words;
- the vocabulary storing a representation of a plurality of reference words;
the system comprising;input means for receiving a time-sequential input pattern representative of a spoken or written word sequence;
a pattern recognizer comprising a word-level matching unit for generating a plurality of sequences of words by statistically comparing the input pattern to the representations of the reference words of the vocabulary;
a cache for storing a plurality of most recently recognized words and word sequence, said cache comprising a plurality of slots, wherein at least one or more slots are reserved for permanent use for storing infrequently retrieved user-designated critical word sequences that are spoken fully and not replaced by subsequently recognized words;
a sequence-level matching unit for selecting a word sequence from the plurality of sequences of words using a statistical language model which provides a probability of a word sequence;
the probability depending on a frequency of occurrence and non-occurrence of the word sequence in the cache.- View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9)
a normalized, non-zero value if the word sequence does not occur in the cache; and
a summation of the normalized value and a frequency-related term which depends on a number of occurrences of the word sequence in the cache, otherwise.
- the vocabulary storing a representation of a plurality of reference words;
-
3. A system as claimed in claim 2, wherein the frequency-related term includes a discounting parameter D which is subtracted from the number of occurrences of the word sequence in the cache.
-
4. A system as claimed in claim 2, wherein the cache is adapted to store the last L recognized word sequences as identifiable word sequences;
- each word sequence being limited to a predetermined sequence length;
the language model specifying the conditional probability of a sequence s of words up to the predetermined sequence length as;
where n(s) is the number of occurrences of the word sequence s in the cache, and γ
is the normalized value.
- each word sequence being limited to a predetermined sequence length;
-
5. A system as claimed in claim 2, wherein the language model specifies the conditional probability of a word wi, given a preceding sequence of words wi−
- 1 . . . wi−
M+1 as;
where n(wi . . . wi−
M+1) is the number of occurrences of the word sequence wi . . . wi−
M+1 in the cache, and γ
(wi . . . wi−
M+1) P(wi|wi−
1 . . . wi−
M+2) is the normalized value.
- 1 . . . wi−
-
6. A system as claimed in claim 5, wherein an end or beginning of a word sequence is represented as a separate unique wordeparator;
- the cache being adapted to store recently recognized word sequences including the word separator.
-
7. A system as claimed in claim 1, wherein the word sequence is at least three.
-
8. A system as claimed in claim 1, wherein the word sequence is four or five.
-
9. A system as claimed in claim 1, wherein a word represents a digit or a command.
Specification