Method and device for recognizing speech

• US 9,514,738 B2
• Filed: 11/13/2012
• Issued: 12/06/2016
• Est. Priority Date: 11/13/2012
• Status: Active Grant

• Current Assignee: Yoichi Ando, Yoshimasa Electronic Inc.
• Original Assignee: Yoichi Ando, Yoshimasa Electronic Inc.
• Inventors: Ando, Yoichi
• Primary Examiner(s): Riley, Marcus T
• Application Number:
  US14/442,191
• Publication Number:
  US 20150348536A1
• Time in Patent Office: 1,484 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G10L 15/01 Assessment or evaluation of speech recognition systems
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/04 Segmentation; Word boundary detection
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates
• CPC Class Codes: G10L 2015/027 Syllables being the recognition units
• CPC Class Codes: G10L 25/06 the extracted parameters being correlation coefficients

Abstract

A speech is recognized using ACF factors extracted from running autocorrelation functions calculated from the speech. The extracted ACF factors are a W_φ(0) (width of ACF amplitude around zero-delay origin), a W_φ(0)max (maximum value of the W_φ(0)), a τ₁ (pitch period), a φ₁ (pitch strength), and a Δφ₁/Δt (rate of the pitch strength change). Syllables in the speech are identified by comparing the ACF factors with templates stored in a database.

Claims

1. A method for recognizing speech, comprising steps of:

recording speech signals by a recording unit;

attenuating, by a LPF, a high frequency component of the speech signals received from the recording unit;

converting, by an AD converter, the speech signals received from the LPF from analog signal to digital signal;

storing, by a storage unit, the speech signals received from the AD converter;

reading out, by a processor, the speech signals stored in the storage unit;

calculating, by the processor, running autocorrelation functions from the speech signals read out from the storage unit;

extracting, by the processor, following ACF factors from the running autocorrelation functions;

a W<sub>φ

(0)</sub> which is a width of ACF amplitude around zero-delay origin,

a W<sub>φ

(0)max</sub> which is a maximum value of the W<sub>φ

(0)</sub>;

a τ

₁ which is a pitch period;

a φ

₁ which is a pitch strength; and

a Δ

φ

₁/Δ

t which is a rate of the pitch strength change;

identifying, by the processor, syllables in the speech signals by comparing the ACF factors with templates stored in a database.

2. The method according to claim 1, further comprising the step of segmenting, by the processor, the speech signals into the syllables based on the ACF factors.

3. The method according to claim 1, further comprising the steps of:

extracting ACF factors from the running autocorrelation functions;

a LL which is a listening level calculated by the amplitude at an origin of non-normalized running ACF;

a LL_max which is a maximum value of the LL; and

a (τ

_e)_min which is a minimum value of effective duration τ

_e; and

segmenting, by the processor, the speech signals into the syllables based on the LL, the (τ

_e)_min, the Δ

φ

₁/Δ

t, the τ

₁ and the W<sub>φ

(0)</sub>.

4. The method according to claim 1, wherein the identifying step is performed in each of the syllables at time points after a (τ

_e)_min.

5. The method according to claim 2, wherein the identifying step is performed in each of the syllables at time points after a (τ

_e)_min.

6. The method according to claim 3, wherein the identifying step is performed in each of the syllables at time points after the (τ

_e)_min.

7. The method according to claim 1, wherein the identifying step identifies the syllables in the speech signals based on a total distance between the ACF factors and the templates.

8. The method according to claim 2, wherein the identifying step identifies the syllables in the speech signals based on a total distance between the ACF factors and the templates.

9. The method according to claim 3, wherein the identifying step identifies the syllables in the speech signals based on a total distance between the ACF factors and the templates.

10. A speech recognition device, comprising:

a recording unit configured to record speech signals;

a LPF configured to attenuate a high frequency component of the speech signals received from the recording unit;

an AD converter configured to convert the speech signals received from the LPF from analog signal to digital signal;

a storage unit configured to store the speech signals received from the AD converter; and

a processor configured to read out the speech signals stored in the storage unit wherein the processor is configured to;

calculate running autocorrelation functions from the speech signals;

extract following ACF factors from the running autocorrelation functions;

a LL which is a listening level calculated by an amplitude at an origin of non-normalized running ACF;

a LL_max which is a maximum value of LL;

a W<sub>φ

(0)</sub> which is a width of ACF amplitude around zero-delay origin,

a W<sub>φ

(0)max</sub> which is a maximum value of the W<sub>φ

(0)</sub>;

a τ

₁ which is a pitch period;

a φ

₁ which is a pitch strength; and

a Δ

φ

₁/Δ

t which is a rate of the pitch strength change; and

identify syllables in the speech signals by comparing the ACF factors with templates stored in a database.

11. The speech recognition device according to claim 10, wherein the processor is further configured to segment the speech signals into syllables based on the ACF factors.

12. The speech recognition device according to claim 10,

wherein the processor further extracts

a LL which is a listening level from the running autocorrelation functions,

a LL_max, which is a maximum value of LL, and

a (τ

_e)_min which is a minimum value of effective duration τ

_e, and

further configured to segment the speech signals into syllables based on the LL, the (τ

_e)_min, the Δ

φ

₁/Δ

t, the τ

₁ and the W<sub>φ

(0)</sub>.

13. The speech recognition device according to claim 10, wherein the processor identifies each of the syllables at a time point after a (τ

_e)_min.

14. The speech recognition device according to claim 11, wherein the processor identifies each of the syllables at a time point after a (τ

_e)_min.

15. The speech recognition device according to claim 12, wherein the processor identifies each of the syllables at a time point after the (τ

_e)_min.

16. The speech recognition device according claim 10, wherein the processor identifies the syllables in the speech signals based on a total distance between the ACF factors and the templates.

17. The speech recognition device according claim 11, wherein the processor identifies the syllables in the speech signals based on a total distance between the ACF factors and the templates.

18. The speech recognition device according claim 12, wherein the processor identifies the syllables in the speech signals based on a total distance between the ACF factors and the templates.