Method of partitioning a sequence of data frames

• US 5,621,848 A
• Filed: 06/06/1994
• Issued: 04/15/1997
• Est. Priority Date: 06/06/1994
• Status: Expired due to Term

• Current Assignee: Google Technology Holdings LLC (Alphabet Inc.)
• Original Assignee: Motorola, Inc. (Motorola Solutions, Inc.)
• Inventors: Wang, Shay-Ping T.
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Sartori, Michael A.
• Application Number:
  US08/254,847
• Time in Patent Office: 1,044 Days
• Field of Search: 395/2.1, 395/2.14, 395/2.18, 395/2.2, 395/2.4, 395/2.55-2.64, 381/41, 381/43, 381/45, 381/49
• US Class Current: 704/211
• CPC Class Codes: G10L 15/04 Segmentation; Word boundary detection
• CPC Class Codes: G10L 15/16 using artificial neural networks
• CPC Class Codes: G10L 25/24 the extracted parameters being the cepstrum

Abstract

A method of analyzing data, particularly a speech signal, first pre-processes the signal by performing analog-to-digital conversion and cepstral analysis, producing a sequence of data frames. Then the sequence of data frames is partitioned into a plurality of data blocks. The data blocks may be subjected to further analysis, for example, by introducing them to a plurality of neural networks. The system may be implemented using either hardware or software or a combination thereof.

Claims

1. In a pattern-recognition system, a method for partitioning a sequence of data frames into a sequence of data blocks, said method comprising the following steps:

(a) converting a spoken word into a signal representing said sequence of data frames;

(b) receiving a first data frame of said sequence of data frames;

(c) equating a current data frame to said first data frame;

(d) equating a current data block to a first data block of said sequence of data blocks;

(e) assigning said current data frame to said current data block;

(f) determining whether there is a next data frame in said sequence of data frames;

(i) if so, proceeding to step (g);

(ii) if not, concluding said method;

(g) equating said current data frame to said next data frame;

(h) equating said current data block to a next data block in said sequence of data blocks;

(j) assigning said current data frame to said current data block;

(k) determining if said current data block is a last one of said sequence of data blocks;

(i) if so, proceeding to step (l);

(ii) if not, returning to step (f); and

(l) equating said next data block to said first data block, and returning to step (j).

2. The method recited in claim 1, further comprising the following step:

including a plurality of data points based on cepstrum coefficients in at lease one of said sequence of data frames.

3. The method recited in claim 1, wherein said method said is implemented using at least one integrated circuit.

4. The method recited in claim 1, wherein said method is implemented using a sequence of computer-executable instructions.

5. In pattern-recognition system, a method for generating a sequence of data blocks, said method comprising the following steps:

(a) converting a spoken word into a signal representing a sequence of data frames;

(b) receiving a first data frame of said sequence of data frames;

(c) equating a current data frame to said first data frame;

(d) equating a current data block to a first data block of said sequence of data blocks;

(e) assigning said current data frame to said current data block;

(f) determining whether there is a next data frame in said sequence of data frames;

(i) if so, proceeding to step (g);

(ii) if not, proceeding to step (m);

(g) equating said current data frame too said next data frame;

(h) equating said current data block to a next data block in said sequence of data blocks;

(j) assigning said current data frame to said current data block;

(k) determining if said current data block is a last one of said sequence of data blocks;

(i) if so, proceeding to step (l);

(ii) if not, returning to step (f);

(l) equating said next data block to said first data block, and returning to step (j); and

(m) determining if said current data block is said last one of said sequence of data blocks;

if so, concluding said method, but if not, copying said current data frame to each of said sequence of data blocks following said current data block.

6. The method recited in claim 5, further comprising the following step:

including a plurality of data points based on cepstrum coefficients in at least one of said sequence of data frames.

7. The method recited in claim 5, wherein said method is implemented using at least one integrated circuit.

8. The method recited in claim 5, wherein said method is implemented using a sequence of computer-executable instructions.

9. In a speech-recognition system, a method for generating a sequence of data blocks, said method comprising the following steps:

(a) converting a spoken word to an electrical signal;

(b) performing an A/D conversion of said electrical signal to produce a digitized signal;

(c) performing cepstral analysis of said digitized signal to produce a sequence of data frames, wherein each of said sequence of data frames includes a plurality of data points based on cepstrum coefficients;

(d) equating a current data frame to a first data frame of said sequence of data frames;

(e) equating a current data block to a first data block of said sequence of data blocks;

(f) assigning said current data frame to said current data block;

(g) determining whether there is a next data frame in said sequence of data frames;

(i) if so, proceeding to step (h);

(ii) if not, concluding said method;

(h) equating said current data frame to said next data frame;

(j) equating said current data block to a next data block in said sequence of data blocks;

(k) assigning said current data frame to said current data block;

(l) determining if said current data block is a last one of said sequence of data blocks;

(i) if so, proceeding to step (m);

(ii) if not, returning to step (g); and

(m) equating said next data block to said first data block, and returning to step (g).

10. The method recited in claim 9, wherein said sequence of frames represents a speech sample.

11. The method recited in claim 9 wherein step (a) comprises the following sub-step:

sampling said spoken word at fixed intervals of time to generate said electrical signal.

12. The method recited in claim 9 wherein each frame in said sequence of frames comprises a plurality of data points, wherein each of said data points represents the value of a function which is based on cepstrum coefficients.

13. The method recited in claim 9, wherein said method is implemented using at least one integrated circuit.

14. The method recited in claim 9, wherein said method is implemented using a sequence of computer-executable instructions.

15. The method recited in claim 9, wherein said sequence of data frames has a finite number of data frames and said sequence of data blocks has a finite length, and said method further comprises the following steps:

determining said finite number of data frames;

defining a number of data frames per data block; and

calculating said finite length of said sequence of data blocks according to said number of data frames and said number of data frames per data block.

16. The method of claim 15, wherein the step of calculating includes the following sub-steps:

dividing said finite number of data frames by said data frames per data block to produce a result; and

rounding said result to a nearest integer greater than said result to produce said finite length of said sequence of data blocks.

17. In a speech-recognition system, a method for generating a sequence of data blocks having a finite length, said method comprising the following steps:

(a) converting a spoken word to an electrical signal;

(b) performing an A/D conversion of said electrical signal to produce a digitized signal;

(c) performing cepstral analysis of said digitized signal to produce a sequence of data frames, wherein said sequence of data frames has a finite number of data frames and each of said sequence of data frames includes a plurality of data points based on cepstrum coefficients, each of the plurality of data points being a 32-bit digital word;

(d) determining said finite number of data frames;

(e) defining a number of data frames per data block;

(f) calculating said finite length of said sequence of data blocks according to said number of data frames and said number of data frames per data block;

(g) equating a current data frame to a first data frame of said sequence of data frames;

(h) equating a current data block to a first data block of said sequence of data blocks;

(j) assigning said current data frame to said current data block;

(k) determining whether there is a next data frame in said sequence of data frames;

(i) if so, proceeding to step (l);

(ii) if not, proceeding to step (g);

(l) equating said current data frame to said next data frame;

(m) equating said current data block to a next data block in said sequence of data blocks;

(n) assigning said current data frame to said current data block;

(o) determining if said current data block is a last one of said sequence of data blocks;

(i) if so, proceeding to step (p);

(ii) if not, returning to step (k); and

(p) equating said next data block to said first data block, and returning to step (k); and

(q) determining if said current data block is said last one of said sequence of data blocks;

if so, concluding said method, but if not, copying said current data frame to each of said sequence of data blocks following said current data block.

18. The method recited in claim 17, wherein step (f) includes the following sub-steps:

dividing said finite number of data frames by said data frames per data block to produce a result; and

rounding said result to a nearest integer greater than said result to generate said finite length of said sequence of data blocks.

19. The method recited in claim 17, wherein step (a) comprises the following sub-step:

sampling said spoken word at fixed intervals of time to generate said electrical signal.

20. The method for partitioning a sequence of data frames recited in claim 17, wherein said method is implemented using at least one integrated circuit.

21. The method recited in claim 17, wherein said method is implemented using a sequence of computer-executable instructions.