METHOD AND DEVICE FOR ASCERTAINING FEATURE VECTORS FROM A SIGNAL

• US 20100017207A1
• Filed: 09/24/2009
• Published: 01/21/2010
• Est. Priority Date: 02/19/2004
• Status: Active Grant

• Current Assignee: Infineon Technologies AG
• Original Assignee: Infineon Technologies AG
• Inventors: Hemmert, Werner, Holmberg, Marcus
• Granted Patent: US 8,064,699 B2
• Time in Patent Office: Days
• US Class Current: 704/236
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence

Abstract

A signal is used to form intermediate feature vectors which are subjected to high-pass filtering. The high-pass-filtered intermediate feature vectors have a respective prescribed addition feature vector added to them.

Claims

1. A method for the computer-aided ascertainment of feature vectors from a digitized signal representing a spoken voice for voice recognition, comprising:

using the signal to form intermediate feature vectors, at least some of whose components indicate a power spectrum from part of the digitized signal;

subjecting the intermediate feature vectors to high-pass filtering by a high-pass filter;

multiplying the intermediate feature vectors by a weighting factor using a weighting unit; and

adding, using an adder, a respective addition feature vector to at least some of the high-pass-filtered intermediate feature vectors, to produce a feature vector representing the spoken voice for use in voice recognition the addition feature vectors used being the respective intermediate feature vectors multiplied by the weighting factor.

2. The method of claim 1, comprising:

performing the high-pass filtering using a non-adaptive high-pass filter.

3. The method of claim 1, comprising:

performing the high-pass filtering using an adaptive high-pass filter.

4. The method of claim 3, comprising:

performing the high-pass filtering using an adaptive wave digital filter.

5. The method of claim 1, comprising:

performing the high-pass filtering using a first-order high-pass filter.

6. The method of claim 1, comprising:

performing the high-pass filtering using a high-pass filter with a cut-off frequency of between 0.2 Hz and 100 Hz.

7. The method of claim 1, comprising:

subtracting a prescribed spectrum signal from the intermediate feature vectors and/or from the high-pass-filtered intermediate feature vectors.

8. The method of claim 7, comprising:

defining the prescribed spectrum signal used to be the first intermediate feature vector and/or the first high-pass-filtered intermediate feature vector.

9. The method of claim 1, comprising:

wherein the weighting factor used is a value between 0.1 and 3.

10. A method for computer-aided voice recognition, comprising:

using a digitized signal representing a spoken voice to form intermediate feature vectors;

subjecting the intermediate feature vectors to high-pass filtering by a high-pass filter;

multiplying the intermediate feature vectors by a weighting factor using a weighting unit;

adding, using an adder, a respective addition feature to at least some of the high-pass-filtered intermediate feature vectors, the addition feature vectors used being the respective intermediate feature vectors multiplied by the weighting factor; and

using the sum formed as feature vectors representing the spoken voice to perform voice recognition.

11. The method of claim 10, comprising:

performing the high-pass filtering using a non-adaptive high-pass filter.

12. The method of claim 10, comprising:

performing the high-pass filtering using an adaptive high-pass filter; and

performing the high-pass filtering using an adaptive wave digital filter.

13. The method of claim 12, comprising:

performing the high-pass filtering using a first-order high-pass filter.

14. The method of claim 10, comprising:

performing the high-pass filtering using a high-pass filter with a cut-off frequency of between 0.5 Hz and 100 Hz.

15. The method of claim 10, comprising:

subtracting a prescribed spectrum signal from the intermediate feature vectors and/or from the high-pass-filtered intermediate feature vectors.

16. The method of claim 15, comprising:

using the first intermediate feature vector for the prescribed spectrum signal.

17. The method of claim 10, comprising:

wherein the weighting factor used is a value between 0.5 and 1.5.