Calibration of speech data acquisition path

• US 20020177998A1
• Filed: 01/18/2002
• Published: 11/28/2002
• Est. Priority Date: 03/28/2001
• Status: Active Grant

• Current Assignee: Texas Instruments, Inc.
• Original Assignee: Texas Instruments, Inc.
• Inventors: Gong, Yifan
• Granted Patent: US 6,912,497 B2
• Time in Patent Office: Days
• US Class Current: 704/226
• CPC Class Codes: G10L 15/065 Adaptation
• 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 method and system for calibration of a data acquisition path is achieved by applying a voice utterance to a first high quality microphone and reference path and to a test acquisition path including a test microphone such as a lower quality one used in a car. The calibration device includes detecting the power density of the reference signal Y_R through the reference path and detecting the power density of the signal Y_N through the acquisition path. A processor processes these signals to provide an output signal representing a noise estimate and channel estimate. The processing uses equation derived by modeling convolutive and additive noise as polynomials with different orders and estimating model parameters using maximum likelihood criterion and simultaneously solving linear equations for the different orders.

Claims

1. A method of calibrating the data acquisition path comprising the steps of:

applying a voice utterance signal simultaneously to two speech transmission paths where the first path is a high quality reference path and a second test path being the acquisition path;

detecting output power density of the reference path to produce signal Y_R and detecting output power density of the second test path to produce the power density of the mismatch signal Y_N;

processing said reference signal Y_R and said mismatch signal Y_N to determine channel estimate and noise estimate of the mismatch using equations derived by modeling convolutive H_Δ

and additive noise N_N as polynomial functions of frequency with one order P for H_Δ

and a different order Q for N_N, estimating model parameters using maximum likelihood criterion to determine the parameter set and simultaneously solving linear equation for both order P and Q.

2. The method of claim 1 wherein said solving step includes solving simultaneously one linear equation for the order of P and another for Q.

3. The method of claim 1 wherein said solving step includes jointly solving linear equation for P+Q variable.

4. A method of calibrating the data acquisition path for each utterance comprising the steps of:

applying a voice utterance signal simultaneously to two speech transmission paths where the first path is a high quality reference path and a second test path being the acquisition path;

detecting output power density of the reference path to produce signal Y_R and detecting output power density of the second test path to produce the power density of the mismatch signal Y_N;

determining for each frame of the utterance the power spectrum for Y_{R and} Y_N;

calculating for all of the frames of the utterance the following

A<sub>(P×

P)</sub>Δ

[A₁,A₂, . . . A_P]^t B<sub>(P×

Q)</sub>Δ

[B₁,B₂, . . . B_P]^t C<sub>(Q×

P)</sub>Δ

[C₁,C₂, . . . C_Q]^t=B^T

D<sub>(Q×

Q)</sub>Δ

[D₁,D₂, . . . D_Q]^t uΔ

[u₁,u₂, . . . u_p]^t with u_p=α

(p−

1,Y_RY_N) vΔ

[υ

₁,υ

₂, . . . υ

_Q]^t with υ

_q=β

(q−

1,Y_N)

for A, B, C, D, u and v;

calculating for the utterance the noise estimate θ

_N using the following;

(D−

B^tA<sup>−

1</sup>B)θ

_N=v−

B^tA<sup>−

1</sup>u;

and calculating for the utterance the channel estimate θ

_H using the following;

θ

_H=A<sup>−

1</sup>(u−

Bθ

_N).

5. A method of calibrating the data acquisition path for each utterance comprising the steps of:

applying a voice utterance signal simultaneously to two speech transmission paths where the first path is a high quality reference path a high and a second test path being the acquisition path to produce the mismatch signal Y_N;

determining for each frame of the utterance the power spectrum for Y_{R and} Y_N;

calculating for all of the frames of each utterance the following

A<sub>(P×

P)</sub>Δ

[A₁,A₂, . . . A_P]^t B<sub>(P×

Q)</sub>Δ

[B₁,B₂, . . . B_P]^t C<sub>(Q×

P)</sub>Δ

[C₁,C₂, . . . C_Q]^t=B^T D<sup>(Q×

Q)</sup>Δ

[D₁,D₂, . . . D_Q]^t uΔ

[u₁,u₂, . . . u_p]^t with u_p=α

(p−

1,Y_RY_N) vΔ

[υ

₁,υ

₂, . . . υ

_Q]^t with υ

_q=β

(q−

1,Y_N)

for A, B, C, D, u and v;

calculating the noise estimate θ

_N and the channel estimate θ

_H using the following;

$\left[\begin{array}{cc}A& B\\ C& D\end{array}\right]<br><br>\text{\hspace{1em}<br><br>}\left[\begin{array}{c}{\mathrm{\theta <br><br>}}_H\\ {\mathrm{\theta <br><br>}}_N\end{array}\right]=\left[\begin{array}{c}u\\ v\end{array}\right].$

6. A data acquisition path calibration device comprising:

a first speech transmission path including a high quality microphone and pre-A/D processing and a second test acquisition path including a lower quality microphone;

means for detecting for each frame the power spectrum density of an input signal received over said first path to produce a reference signal Y_R and for detecting for each frame the power spectrum density of said input signal received over said acquisition path to produce mismatch signal Y_N

means for determining the noise estimate and the channel estimate by calculating for each utterance;

A<sub>(P×

P)</sub>Δ

[A₁,A₂, . . . A_P]^t B<sub>(P×

Q)</sub>Δ

[B₁,B₂, . . . B_P]^t C<sub>(Q×

P)</sub>Δ

[C₁,C₂, . . . C_Q]^t=B^T D<sub>(Q×

Q)</sub>Δ

[D₁,D₂, . . . D_Q]^t uΔ

[u₁,u₂, . . . u_p]^t with u_p=α

(p−

¹,Y_RY_N) vΔ

[υ

₁,υ

₂, . . . υ

_Q]^t with υ

_q=β

(q−

1,Y_N)

for A, B, C, D, u and v;

calculating for the utterance the noise estimate θ

_N using the following;

(D−

B^tA<sup>−

1</sup>B)θ

_N=v−

B^tA<sup>−

1</sup>u;

and calculating for the utterance the channel estimate θ

_H using the following;

θ

_H=A<sup>−

1</sup>(u−

Bθ

_N).

7. A data acquisition path calibration device comprising:

a first speech transmission path including a high quality microphone and pre-A/D processing and a second test acquisition path including a lower quality microphone;

means for detecting for each frame the power spectrum density of an input signal received over said first path to produce a reference signal Y_R and for detecting for each frame the power spectrum density of said input signal received over said acquisition path to produce mismatch signal Y_N

means for determining the noise estimate and the channel estimate by calculating for each utterance;

A<sub>(P×

P)</sub>Δ

[A₁,A₂, . . . A_P]^t B<sub>(P×

Q)</sub>Δ

[B₁,B₂, . . . B_P]^tC<sub>(Q×

P)</sub>Δ

[C₁,C₂, . . . C_Q]^t=B^T D<sub>(Q×

Q)</sub>Δ

[D₁,D₂, . . . D_Q]^t uΔ

[u₁,u₂, . . . u₉]^t with u_p=α

(p−

1,Y_RY_N) vΔ

[υ

₁,υ

₂, . . . υ

_Q]^t with υ

_q=β

(q−

1,Y_N)

for A, B, C, D, u and v;

calculating the noise estimate θ

_N and the channel estimate θ

_H using the following;

$\left[\begin{array}{cc}A& B\\ C& D\end{array}\right]<br><br>\text{\hspace{1em}<br><br>}\left[\begin{array}{c}{\mathrm{\theta <br><br>}}_H\\ {\mathrm{\theta <br><br>}}_N\end{array}\right]=\left[\begin{array}{c}u\\ v\end{array}\right].$