AU7145600A

AU7145600A - Method and apparatus for estimating a spectral model of a signal used to enhance a narrowband signal

Info

Publication number: AU7145600A
Application number: AU71456/00A
Authority: AU
Inventors: Julien Roger Epps; Warwick Harvey Holmes; Mark Mclean Thomson
Original assignee: University of New South Wales; Motorola Australia Pty Ltd
Current assignee: University of New South Wales; Motorola Solutions Australia Pty Ltd
Priority date: 1999-11-12
Filing date: 2000-11-06
Publication date: 2001-05-17
Anticipated expiration: 2020-11-06
Also published as: AU754612B2

Description

P/00/011 Regulation 3.2

AUSTRALIA

Patents Act 1990

ORIGINAL

COMPLETE SPECIFICATION STANDARD PATENT Invention Title: "METHOD AND APPARATUS FOR ESTIMATING A SPECTRAL MODEL OF A SIGNAL USED TO ENHANCE A NARROWBAND SIGNAL" The following statement is a full description of this invention, including the best method of performing it known to us: CR1057AC METHOD AND APPARATUS FOR ESTIMATING A SPECTRAL MODEL OF A SIGNAL USED TO ENHANCE A NARROWBAND

SIGNAL

This invention relates to a method and apparatus for estimating a spectral model of a signal used to enhance a narrowband signal, especially, though not exclusively, so as to improve the performance of a class of methods S *10 used for enhancing the quality of narrowband speech signals.

"BACKGROUND OF THE INVENTION The power spectral density, or power spectrum, of a signal describes the 15 proportion of signal power associated with different ranges of frequency. Thus the integral of this function with respect to frequency over some range gives i the relative amount of power within that range. Speech signals in many voice ooooo S"communication systems have most of their significant power associated with a bandwidth, or frequency range, of approximately 0.3 4 kHz. This is a 20 relatively narrow bandwidth compared with the range of frequencies present in acoustic speech signals, and compared with the range that can be perceived by the human auditory system. Such a signal is therefore often referred to as a narrowband signal. To improve the perceived quality of such a signal, it is desirable to generate from it an enhanced signal having a larger bandwidth.

Such a signal is referred to as a wideband signal.

In some systems, the enhanced signal consists of one component that is equal, or approximately equal, to the narrowband signal, and a second component with significant power associated with frequencies outside the bandwidth of this signal. The addition of the second component to the narrowband signal is intended to create a wideband signal having improved perceived quality.

2 A number of methods of generating the second component are known, such as that described in "Bandwidth enhancement of narrowband speech signals", by H. Carl and U. Heute, in Signal Processing VII, Theory and Appicalions., EUSIPCO, 1994, vol. 2, pp. 1178-1181. Common to many methods is a need to determine a model of the spectrum of the second component. The term spectrum model is here used to mean a function of frequency that has some features that are characteristic of the power spectrum of the signal. For example, a spectral envelope, which is a smoothed approximation to a signal's power spectrum is a spectral model. The power 10 spectrum itself may also be regarded as a spectral model. A spectral model of S:I" the second component is calculated from information about the narrowband ooo.oi signal, most commonly from the spectral envelope of the narrowband signal.

In this specification, including the claims, the terms "comprises", "comprising" or similar terms are intended to mean a non-exclusive inclusion, such that a method or apparatus that comprises a list of elements does not include those elements solely, but may well include other elements not listed.

S"BRIEF SUMMARY OF THE INVENTION 20 It is therefore an object of the present invention to provide a method and apparatus for determining parameters of a spectral model of a signal used to enhance a narrowband signal.

Accordingly, in one aspect, the invention provides a method of determining parameters of a spectral model of a signal used to enhance a narrowband signal, the method comprising the steps of: receiving a first set of parameters representing a first spectral model, wherein the first spectral model is a function of frequency; determining a second set of parameters representing a second spectral model, wherein the second spectral model is a function of frequency that approximates the first spectral model over a first range of; determining a third set of parameters representing a spectrum correction function from the first set of parameters, wherein the spectrum correction function is a function of frequency; and determining a fourth set of parameters representing a third spectral model wherein for any particular frequency within a second range of frequencies, the third spectral model depends on values of both the second spectral model and the spectrum correction function at the particular frequency.

Preferably, the third spectral model may provide the spectral model of 10 the signal used to enhance the narrowband signal.

Suitably, the step of estimating a spectrum correction function may .•o.oi comprise the step of determining a set of quantised parameters, wherein the set of quantised parameters is one of a plurality of predetermined sets of oooo codebook input parameters.

Preferably, each of the plurality of predetermined sets of codebook o:ooo input parameters may represent a corresponding narrowband codebook •oooo spectrum.

Suitably, the narrowband codebook spectrum associated with the set of S 20 quantised parameters may approximate the first spectral model.

Preferably, the spectrum correction function may be one of a plurality of predetermined functions of frequency each represented by a set of codebook output parameters, and each of the predetermined functions of frequency represented by a set of codebook output parameters is associated with a predetermined set of codebook input parameters.

According to a second aspect, the invention provides an apparatus for estimating a spectral envelope of a signal used to enhance a narrowband signal, the apparatus comprising: an input terminal for receiving a first set of parameters representing a first spectral model, wherein the first spectral model is a function of frequency; an analyser having an input coupled to the input terminal and an output, for determining a second set of parameters representing a second spectral model, wherein the second spectral model is a function of frequency that approximates the first spectral model over a first range of frequencies; a transformer having an input coupled to the output of the analyser and an output, for determining a third set of parameters representing a spectrum correction function from the first set of parameters, wherein a spectrum correction function is a function of 10 frequency; a combiner having a first input coupled to the output of the ooooo transformer, a second input coupled to the output of the analyser and an output, for determining a fourth set of parameters representing a third spectral model, wherein for any particular frequency within a second S "15 range of frequencies, the third spectral model depends on values of both the second spectral model and the spectrum correction function at the .o.o.i particular frequency.

Preferably, the transformer used for determining a difference spectrum from parameters of the first spectral model may comprise: 20 a first codebook for determining a set of quantised parameters, wherein the set of quantised parameters is one of a plurality of predetermined sets of codebook input parameters, and wherein each of the plurality of predetermined set of codebook input parameters represents a corresponding narrowband codebook spectrum.

Suitably, the narrowband codebook spectrum associated with the set of quantised parameters may approximate the first spectral model better.

Preferably, there may be a second codebook for determining the spectrum correction function, the spectrum correction function being one of a plurality of predetermined functions of frequency each represented by a set of codebook output parameters, and wherein each of the predetermined functions of frequency represented by a set of codebook output parameters is associated with a predetermined set of codebook input parameters.

BRIEF DESCRIPTION OF THE DRAWINGS An embodiment of the invention will now be more fully described, by way of example, with reference to the drawings, of which: FIG. 1 shows a block diagram of an apparatus according to one embodiment of the invention for estimating a spectral model of a signal used to enhance a narrowband signal.

FIG. 2 shows a block diagram of a transformer used in the apparatus of FIG. 1 for transforming parameters of a narrowband spectral envelope into a highband difference spectrum.

15 DETAILED DESCRIPTION OF THE DRAWINGS Thus, the overall scheme of the invention according to one embodiment °ooo° is shown in FIG. 1 and FIG. 2.

In this embodiment, a first spectral model is a spectral envelope defined over a range of frequencies from 0 4 kHz. This is a narrowband range. The spectral envelope is represented by parameters comprising linear prediction parameters and a gain parameter. As shown in FIG. 1, a received first set of spectral parameters is conveyed from an input terminal 100 to an analyser 102, which produces a second set of parameters corresponding to a second spectral model. The second spectral model is preferably a wideband spectral envelope covering a first range of frequencies typically covering frequencies from 0 8 kHz, but may cover a different frequency range.

In one preferred embodiment, the second spectral model is a function of frequency,f, having the form 1 P(f) Xkobk (27f) over the frequency range from 0 8 kHz. The analyser 102 determines estimates of parameters, bk, such that P(f) is as close as possible to the first spectral model in the least squares sense over a frequency range from 0 4 kHz. The estimates of the parameters are determined using a non-linear optimisation method such as a gradient descent algorithm.

It will be appreciated that other parametric functions could also used to represent the second spectral model. for example, in another embodiment, the first set of parameters includes a parameter representing the degree of voicing of the signal represented by the first spectral model, and the second spectral model is chosen such that the log of the second spectral model is a straight line whose gradient is dependent upon the parameter representing the degree of voicing of the signal represented by the first spectral model.

As shown in FIG. 1, the first set of parameters is also conveyed from input terminal 100 to a transformer 101 which produces a spectrum correction function used for determining a third set of parameters. The third set of parameters represents the spectrum correction function derived from the first set of parameters. The spectrum correction function is a function of frequency defined preferably over a range of frequencies from 4 8 kHz. This is referred *to as a highband range.

20 As shown in FIG. 2, the transformer comprises a converter 200 which first converts the linear prediction (LP) parameters representing the first spectral envelope to another representation, such as line spectral frequencies, using any desired procedure, for example, such as that described in Discrete- Time Processing of Speech Signals, by J. R. Deller Jr, J. G. Proakis and J. H. L.

Hansen, Macmillan, 1993.

The transformer includes two codebooks, a first codebook 201 and a second codebook 202. The first codebook 201 comprises predetermined sets of codebook input parameters, which may be line spectral frequency, or other, parameters corresponding to narrowband spectral envelopes, and the second codebook 202 comprises predetermined sets of codebook output parameters, which are samples of possible highband spectrum correction functions. Each set of line spectral frequency parameters contained within the first codebook 201 and each possible spectrum correction function in the second codebook 202 is identified by an index. Each set of line spectral frequency parameters in the first codebook 201 is associated uniquely with the spectrum correction function in the second codebook 202 having the same index. A set of quantised parameters are chosen from among the sets of line spectral frequency parameters contained in the first codebook 201. This is achieved by converting the LP and gain parameters received at the input terminal 100 into line spectral frequency parameters, and choosing the predefined set of line spectral 10 frequency parameters that are closest to those provided by the converter as measured using a Euclidean distance measure The corresponding spectrum correction function in the second codebook 202 is output from the transformer 101.

In another possible embodiment, the codebook input parameters may 15 represent possible differences between the first spectral model and the third spectral model over the range from 0 4 kHz. In this case, the possible ooooo S°difference closest to the actual difference between the first and third spectral ooooo models is identified, and the corresponding spectrum correction function in the second codebook is output from the transformer.

It will be appreciated that although codebook mapping has been described for the transformer in this embodiment, various other methods could alternatively be employed.

Prior to using the system, the codebooks 201 and 202 must be trained using examples of codebook input parameters and codebook output parameters.

Training of the codebooks can be performed using any desired procedure, for example, utilising the algorithm described in "Bandwidth enhancement of narrowband speech signal", by H. Carl and U. Heute, in Signal Processing VII, Theory and Applications, EUSIPCO, 1994, vol. 2, pp. 1178-1181.

Values of the second spectral model is combined with the first spectral model and the spectrum correction function using a combiner 103 to produce a third spectral model that is represented by a fourth set of parameters. The 8 combiner 103 has three inputs coupled, respectively to the input terminal 100, the transformer 101 and the analyser 102. Over the range from 4 kHz to 8 kHz the third spectral model is the sum of the second spectral model and the spectrum correction function. In one embodiment, in a small range of frequencies below 4 kHz, the third spectral model may be interpolated between the first spectral model and the sum of the second spectral model and the spectrum correction function. For example, the third spectral model may be smoothly interpolated between the first spectral model at 3.8 kHzi-I and the sum of the second spectral model and the spectrum correction function at 4 kHz.

Advantageously, the invention provides for an apparatus for estimating a spectral envelope of a signal used to enhance a narrowband signal. The invention also provides for a method of determining a spectral model of a .'.signal used to enhance a narrowband signal. The method includes receiving 15 the first set of spectral parameters representing the first spectral model which is a spectral envelope and therefore it is a function of frequency. The method

SSSSSS

then effects a step of determining the second set of spectral parameters 555555 representing the second spectral model which is a function of frequency that approximates to the first spectral model over the first range of frequencies. A 5555 20 step of determining the third set of parameters is then effected followed by determining the fourth set of parameters representing the third spectral model.

It will be appreciated that although one particular embodiment has been described here in detail, various modifications and improvements can be made by a person skilled in the art without departing from the scope of the present invention.

w~i-nt ode with this application

Claims

1. A method of determining parameters of a model of the spectrum of a digitised waveform, the method comprising the steps of: receiving a first input signal; determining a first power spectrum estimate of the first input signal, wherein the power spectrum estimate is a function of frequency; ~computing a masking function, wherein the masking function is a function of frequency that depends on the first power spectrum estimate; modifying the first power spectrum estimate to produce a second power spectrum estimate, wherein a value of the second power spectrum estimate is less than a value of the first power spectrum estimate at frequencies at which a value of S-the masking function is greater than the value of the first power spectrum estimate; and determining a set of filter parameters, wherein a squared magnitude of a frequency response associated with a filter controlled by the set of filter parameters is an approximation to the second power spectrum estimate.

2. A method as claimed in claim 1, wherein in the step of determining the first power spectrum estimate includes calculating squared magnitudes of Fast Fourier Transforms of the first input signal.

3. A method as claimed in claim 1, wherein in the step of determining a set of filter parameters includes calculating an autocorrelation function from the second power spectrum estimate.

4. A method as claimed in claim 3, wherein in the filter parameters are for an autoregressive filter. 11 range of frequencies, the third spectral model depends on values of both the second spectral model and the spectrum correction function at the particular frequency.

8. An apparatus as claimed in claim 6, wherein the transformer used for determining a difference spectrum from parameters of the first spectral model comprises: a first codebook for determining a set of quantised parameters, wherein the set of quantised parameters is one of a plurality of 10 predetermined sets of codebook input parameters, and wherein each of the plurality of predetermined set of codebook input parameters represents a corresponding narrowband codebook spectrum.

9. An apparatus as claimed in claim 7, wherein the narrowband codebook 15 spectrum associated with the set of quantised parameters approximates the first spectral model better.

10. An apparatus as claimed in claim 8, wherein there is a second codebook for determining the spectrum correction function, the spectrum correction 20 function being one of a plurality of predetermined functions of frequency each represented by a set of codebook output parameters, and wherein each of the predetermined functions of frequency represented by a set of codebook output parameters is associated with a predetermined set of codebook input parameters.

11. A method of determining parameters of a spectral model of a signal used to enhance a narrow band signal substantially as hereinbefore described with reference to the accompanying drawings. 12

12. An apparatus for estimating a spectral envelope of a signal used to enhance a narrowband signal substantially as hereinbefore described with reference to the accompanying drawings. o