Method and apparatus for adaptive speech detection by applying a probabilistic description to the classification and tracking of signal components

• US 6,691,087 B2
• Filed: 09/30/1998
• Issued: 02/10/2004
• Est. Priority Date: 11/21/1997
• Status: Expired due to Term

• Current Assignee: LG Electronics, Inc. (LG Corporation), SRI International, Inc.
• Original Assignee: LG Electronics, Inc. (LG Corporation), Sarnoff Corporation (SRI International, Inc.)
• Inventors: Parra, Lucas, de Vries, Aalbert
• Primary Examiner(s): Chawan, Vijay
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US09/163,697
• Publication Number:
  US 20020184014A1
• Time in Patent Office: 1,959 Days
• Field of Search: 704/331, 704/236, 704/240
• US Class Current: 704/240
• CPC Class Codes: G10L 2025/786 Adaptive threshold
• CPC Class Codes: G10L 25/78 Detection of presence or absence of voice signals switching of direction of transmission by voice frequency in two-way loud-speaking telephone systems H04M9/10

Abstract

A signal processing system for detecting the presence of a desired signal component by applying a probabilistic description to the classification and tracking of various signal components (e.g., desired versus non-desired signal components) in an input signal is disclosed.

Claims

1. A signal processing method for detecting a presence of a desired signal component from an input signal having more than one signal component, said method comprising the steps of:

a) applying a windowing function to the input signal to generate a plurality of frames;

b) selecting at least one feature for processing said plurality of frames; and

c) detecting the presence of the desired signal component in said frames in accordance with said selected feature by categorizing said frames using a probabilistic description, wherein said detecting step (c) employs an Expectation-Maximization method having a probabilistic description of p(y)=m₁N(y ;

μ

₁,Σ

₁)+m₂N(y ;

μ

₂,Σ

₂), wherein said probabilistic description is optimized in a single pass.

2. The method of claim 1, wherein said detecting step (c) employs a modified Expectation-Maximization (EM) method.

3. The method of claim 2, wherein said detecting step (c) employs said modified Expectation-Maximization (EM) having the following parameters:

$z_i^{\left(k\right)}<br><br>\left(k+1\right)=\frac{m_i<br><br>\left(k\right)<br><br>N<br><br>\left(y<br><br>\left(k+1\right);<br><br>{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right),\underset{i}{\sum <br><br>}<br><br>\left(k\right)\right)}{\underset{i}{\sum <br><br>}<br><br>m_i<br><br>\left(k\right)<br><br>N<br><br>\left(y<br><br>\left(k+1\right);<br><br>{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right),\underset{i}{\sum <br><br>}<br><br>\left(k\right)\right)};<br><br>$$w<br><br>\left(k\right)=\underset{i}{\sum <br><br>}<br><br>v_i<br><br>\left(k\right);<br><br>$ 

v_i(k+1)=β

(k)v_i(k)+z_i^(k)(k+1);

4. The method of claim 3, wherein said detecting step (c) employs said modified Expectation-Maximization (EM) having the following forgetting factor $\beta$

=1-2

 





zi

(k)-mi

(k)

N.

5. The method of claim 1, wherein said detecting step (c) detects the presence of the desired signal component that is a speech component.

6. A signal processing apparatus for detecting a presence of a desired signal component from an input signal having more than one signal component, said apparatus comprising:

a windowing module for applying a windowing function to the input signal to generate a plurality of frames;

a feature selection module for selecting at least one feature for processing said plurality of frames; and

a detection module for detecting the presence of the desired signal component in said frames in accordance with said selected feature by categorizing said frames using a probabilistic description, wherein said probabilistic description employs an Expectation-Maximization (EM) method, wherein said probabilistic description is p(y)=m₁N(y ;

μ

,Σ

₁)+m₂N(y′

,μ

₂,Σ

₂), wherein said probabilistic description is optimized in a single pass.

7. The apparatus of claim 6, wherein said probabilistic description employs a modified Expectation-Maximization (EM) method.

8. The apparatus of claim 7, wherein said modified Expectation-Maximization (EM) has the following parameters:

$z_i^{\left(k\right)}<br><br>\left(k+1\right)=\frac{m_i<br><br>\left(k\right)<br><br>N<br><br>\left(y<br><br>\left(k+1\right);<br><br>{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right),\underset{i}{\sum <br><br>}<br><br>\left(k\right)\right)}{\underset{i}{\sum <br><br>}<br><br>m_i<br><br>\left(k\right)<br><br>N<br><br>\left(y<br><br>\left(k+1\right);<br><br>{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right),\underset{i}{\sum <br><br>}<br><br>\left(k\right)\right)};<br><br>$$w<br><br>\left(k\right)=\underset{i}{\sum <br><br>}<br><br>v_i<br><br>\left(k\right);<br><br>$$v_i<br><br>\left(k+1\right)=\mathrm{\beta <br><br>}<br><br>\left(k\right)<br><br>v_i<br><br>\left(k\right)+z_i^{\left(k\right)}<br><br>\left(k+1\right);<br><br>$$\begin{array}{c}{m}_i<br><br>\left(k+1\right)=\frac{1}{w<br><br>\left(k+1\right)}<br><br>\left(\mathrm{\beta <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>\left(k\right)<br><br>w<br><br>\left(k\right)<br><br>m_i<br><br>\left(k\right)+z_i^{\left(k\right)}<br><br>\left(k+1\right)\right);<br><br>\\ {\mathrm{\mu <br><br>}}_i<br><br>\left(k+1\right)=\frac{1}{v_i<br><br>k+1)}<br><br>\left(\mathrm{\beta <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>\left(k\right)<br><br>v_i<br><br>\left(k\right)<br><br>{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right)+z_i^{\left(k\right)}<br><br>\left(k+1\right)<br><br>y<br><br>\left(k\right)\right);<br><br>\text{\hspace{1em}<br><br>}<br><br>\mathrm{and}\\ \underset{i}{\sum <br><br>}<br><br>\left(k+1\right)=\frac{1}{v_i<br><br>\left(k+1\right)}<br><br>\left(\mathrm{\beta <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>\left(k\right)<br><br>v_i<br><br>\left(k\right)<br><br>\underset{i}{\sum <br><br>}<br><br>\left(k\right)+z_i^{\left(k\right)}<br><br>\left(k+1\right)<br><br>\left(y<br><br>\left(k+1\right)-{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right)\right)<br><br>{\left(y<br><br>\left(k+1\right)-{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right)\right)}^T\right).\end{array}$

9. The apparatus of claim 8, wherein said modified Expectation-Maximization (EM) has a forgetting factor $\beta$

=1-2



zi

(k)-mi

(k)

N.

10. The apparatus of claim 6, wherein said desired signal component that is a speech component.

11. A computer-readable medium having stored thereon a plurality of instructions, the plurality of instructions including instructions which, when executed by a processor, cause the processor to perform the steps comprising of:

a) applying a windowing function to the input signal to generate a plurality of frames;

b) selecting at least one feature for processing said plurality of frames; and

c) detecting the presence of the desired signal component in said frames in accordance with said selected feature by categorizing said frames using a probabilistic description, wherein said detecting step (c) employs an Expectation-Maximization method having a probabilistic description of p(y)=m₁N(y′

,μ

₁,Σ

₁)+m₂N(y′

,μ

₂,Σ

₂), wherein said probabilistic description is optimized in a single pass.

12. The computer-readable medium of claim 11, wherein said detecting step (c) employs a modified Expectation-Maximization (EM) method.

13. The computer-readable medium of claim 12, wherein said detecting step (c) employs said modified Expectation-Maximization (EM) having the following parameters:

$z_i^{\left(k\right)}<br><br>\left(k+1\right)=\frac{m_i<br><br>\left(k\right)<br><br>N<br><br>\left(y<br><br>\left(k+1\right);<br><br>{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right),\underset{i}{\sum <br><br>}<br><br>\left(k\right)\right)}{\underset{i}{\sum <br><br>}<br><br>m_i<br><br>\left(k\right)<br><br>N<br><br>\left(y<br><br>\left(k+1\right);<br><br>{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right),\underset{i}{\sum <br><br>}<br><br>\left(k\right)\right)};<br><br>$$w<br><br>\left(k\right)=\underset{i}{\sum <br><br>}<br><br>v_i<br><br>\left(k\right);<br><br>$$v_i<br><br>\left(k+1\right)=\mathrm{\beta <br><br>}<br><br>\left(k\right)<br><br>v_i<br><br>\left(k\right)+z_i^{\left(k\right)}<br><br>\left(k+1\right);<br><br>$$\begin{array}{c}{m}_i<br><br>\left(k+1\right)=\frac{1}{w<br><br>\left(k+1\right)}<br><br>\left(\mathrm{\beta <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>\left(k\right)<br><br>w<br><br>\left(k\right)<br><br>m_i<br><br>\left(k\right)+z_i^{\left(k\right)}<br><br>\left(k+1\right)\right);<br><br>\\ {\mathrm{\mu <br><br>}}_i<br><br>\left(k+1\right)=\frac{1}{v_i<br><br>k+1)}<br><br>\left(\mathrm{\beta <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>\left(k\right)<br><br>v_i<br><br>\left(k\right)<br><br>{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right)+z_i^{\left(k\right)}<br><br>\left(k+1\right)<br><br>y<br><br>\left(k\right)\right);<br><br>\text{\hspace{1em}<br><br>}<br><br>\mathrm{and}\\ \underset{i}{\sum <br><br>}<br><br>\left(k+1\right)=\frac{1}{v_i<br><br>\left(k+1\right)}<br><br>\left(\mathrm{\beta <br><br>}<br><br>\text{\hspace{1em}<br><br>}<br><br>\left(k\right)<br><br>v_i<br><br>\left(k\right)<br><br>\underset{i}{\sum <br><br>}<br><br>\left(k\right)+z_i^{\left(k\right)}<br><br>\left(k+1\right)<br><br>\left(y<br><br>\left(k+1\right)-{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right)\right)<br><br>{\left(y<br><br>\left(k+1\right)-{\mathrm{\mu <br><br>}}_i<br><br>\left(k\right)\right)}^T\right).\end{array}$

14. The computer-readable medium of claim 13, wherein said detecting step (c) employs said modified Expectation-Maximization (EM) having the following forgetting factor $\beta$

=1-2



zi

(k)-mi

(k)

N.