Method and apparatus for improving acoustic fast match speed using a cache for phone probabilities

• US 5,963,905 A
• Filed: 10/24/1997
• Issued: 10/05/1999
• Est. Priority Date: 10/24/1997
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Novak, Miroslav, Picheny, Michael Alan
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): ABEBE, DANIEL DEMELASH
• Application Number:
  US08/957,200
• Time in Patent Office: 711 Days
• Field of Search: 704/256, 704/254, 704/243
• US Class Current: 704/254
• CPC Class Codes: G10L 15/14 using statistical models, e.g. Hidden Markov Models [HMMs] G10L15/18 takes precedence
• CPC Class Codes: G10L 2015/081 Search algorithms, e.g. Baum-Welch or Viterbi

Abstract

Methods and apparatus for performing a tree search based acoustic fast match in a speech recognition system for decoding a speech utterance, the tree having a tree root and tree nodes connected by tree branches, the tree nodes having phonetic models associated therewith, are provided. An illustrative embodiment of the method comprises: providing a cache having cache cells for storing phone probabilities therein; selecting a first branch leading to a next node, said branch selection starting at the tree root; accessing the cache to select a particular cache cell where the probability of a particular match is stored; evaluating the phonetic model to obtain the probability and storing the probability and an associated end time in the cache cell, if the cache cell accessed in the accessing step does not contain the required probability; using the probability value and the associated end time stored in the cache cell, if the cache cell accessed in the accessing step contains the required probability; selecting a new branch to proceed to the next node; and iteratively continuing from the accessing step until the whole tree is traversed and all possible word candidates associated with the speech recognition system are evaluated.

Claims

1. A method for performing a tree search based acoustic fast match in a speech recognition system for decoding a speech utterance, the tree having a tree root and tree nodes connected by tree branches, the tree nodes having phonetic models associated therewith, the method comprising the steps of:

(a) providing a cache having cache cells for storing phone probabilities therein;

(b) selecting a first branch leading to a next node, said branch selection starting at the tree root;

(c) accessing the cache to select a particular cache cell where the probability of a particular match is stored;

(d) evaluating the phonetic model to obtain the probability and storing the probability and an associated end time in the cache cell, if the cache cell accessed in the accessing step does not contain the required probability;

(e) using the probability value and the associated end time stored in the cache cell, if the cache cell accessed in the accessing step contains the required probability;

(f) selecting a new branch to proceed to the next node; and

(g) iteratively continuing from the accessing step until the tree is traversed and the word candidates associated with the speech recognition system are evaluated.

2. The method of claim 1, wherein step (g) further includes traversing the whole tree and evaluating all possible word candidates.

3. The method of claim 1, wherein the cache is accessed as a two-dimensional array.

4. The method of claim 1, wherein the cache is accessed as a three-dimensional array.

5. The method of claim 4, wherein the cache cell is selected in the accessing step based on the identity of a corresponding current phone, the identity of a corresponding previous phone, and a corresponding end time of the previous phone.

6. The method of claim 5, wherein the corresponding current phone identity of the current phone is associated with the next node of step (b).

7. The method of claim 5, wherein the corresponding previous phone identity is associated with a corresponding previous node.

8. The method of claim 5, wherein combinations of current phones and previous phones are mapped into the cache cells.

9. The method of claim 1, wherein the step of accessing the cache comprises:

mapping an identity of a current phone and an identity of a previous phone into a di-phone index;

determining if, for a certain di-phone, an index exists; and

quantizing and mapping an end time value.

10. The method of claim 9, wherein the accessing step further includes inserting an earlier access time into a cache cell to indicate that said cell contains a probability relevant to a subsequent access time.

11. The method of claim 9, wherein the accessing step further comprises discarding a probability stored in a cell when a selected cell indicates an earlier access time.

12. An apparatus for performing a tree search based acoustic fast match, the tree having a tree root and tree nodes connected by tree branches, the tree nodes having phonetic models associated therewith, the apparatus comprising:

(a) a cache having cache cells for storing phone probabilities therein;

(b) means for selecting a first branch leading to a next node, said branch selection starting at the tree root;

(c) means for accessing the cache to select a particular cache cell where the probability of a particular match is stored;

(d) means for evaluating the phonetic model to obtain the probability and storing the probability and an associated end time in the cache cell, if the cache cell accessed in the accessing step does not contain the required probability;

(e) means for using the probability value and the associated end time stored in the cache cell, if the cache cell accessed in the accessing step contains the required probability; and

(f) means for selecting a new branch to proceed to the next node and iteratively continuing from the accessing function of the accessing means until the tree is traversed and the word candidates associated with the speech recognition system are evaluated.

13. The apparatus of claim 12, wherein an earlier access time is inserted into a cache cell to indicate that said cell contains a probability relevant to a subsequent access time.

14. The apparatus of claim 12, further comprising means for discarding a probability stored in a cell when a selected cell indicates an earlier access time.

15. An apparatus for performing a tree search based acoustic fast match, comprising:

means for performing an iterative search in a local time region;

a cache having cache cells for storing phone probabilities therein;

means for mapping an identity of a current phone and an identity of a previous phone into a di-phone index;

means for quantizing and mapping an end time value;

means for determining if, for a certain di-phone, an index exists; and

means for selecting a cache cell with respect to a particular end time value if, for a certain di-phone, an index exists.

16. The apparatus of claim 15, wherein an earlier access time is inserted into a cache cell to indicate that said cell contains a probability relevant to a subsequent access time.

17. The apparatus of claim 15, further comprising means for discarding a probability stored in a cell when a selected cell indicates an earlier access time.

18. A method for improving tree search based acoustic fast match speed using a cache for phone probabilities, comprising the steps of:

(a) inputting the identity of a current phone, the identity of a previous phone, and the end time of the previous phone;

(b) mapping a cache cell;

(c) determining whether a cell exists;

(d) evaluating a phone model, if the cell determined at step (c) does not exist;

(e) determining whether the value is valid, if the cell determined at step (c) exists;

(f) using a cache value determined at step (e), if the value determined at step (e) is valid;

(g) determining whether the previous phone was computed, if the value determined at step (e) is not valid;

(h) evaluating the phone model and storing the results in cache, if the previous phone was computed at step (g);

(i) moving one phone backward, if the previous phone determined at step (g) was not computed;

(j) determining whether the previous phone was computed;

(k) going back to step (i), if the previous phone determined at step (j) was not computed;

(l) evaluating the phone model and moving one phone forward, if the previous phone determined at step (j) was computed;

(m) determining whether the one move forward is back to the current phone;

(n) going back to step (l), if the one move forward is not back to the current phone as determined in step (m); and

(o) evaluating the phone model and storing the results in cache, if back to the current phone as determined in step (m).