KR950003557B1 - Encoding method of voice sample and signal sample - Google Patents

Abstract

No content.

Description

[Name of invention]

Speech Sample and Signal Sample Encoding Methods

[Brief Description of Drawings]

1 is a block diagram illustrating the present invention.

2 is a simple vector diagram illustrating an aspect of the present invention.

Detailed description of the invention

[Technical Field]

FIELD OF THE INVENTION The present invention relates generally to speech coders, and more particularly to digital speech coders using vector excitation sources.

[Background of invention]

Speech coders are known in the art. Some speech coders convert analog speech samples into digital representations and then use linear predictive coding to represent spectral speech information. Other speech coders improve the conventional linear predictive coding technique by providing an excitation signal relative to the original speech signal. Already issued US Patent No. 4,817,157 describes a digital speech coder with an improved vector excitation source, in which the codebook of the excitation vector is approximated to the original signal by being accessed to select the excitation signal that best matches the utilization information. It provides a reconstructed speech signal expressing.

In general, the final decoded speech signal represents the original unencoded speech signal when there is a certain number of preliminary excitation vectors used as excitation sources. However, increasing performance in this manner generally increases the codebook size, typically complicates processing and increases data rates.

Therefore, there is a need for a digital speech coder using a vector excitation signal, where a codebook of a predetermined size maximizes the decoded speech signal quality with little increase in data rate while minimizing complexity.

[Summary of invention]

These needs and others can be solved with a digital speech coder facility using vector excitation sources with improved sound quality. According to the present invention, when encoding speech samples, the encoder first determines the pitch period parameter for the speech samples. In part according to this pitch period parameter, a particular coded excitation signal can be determined irrespective of the pitch filter coefficient, whereby the pitch filter coefficient parameter optimizes a particular speech sample. This methodology can account for preliminary excitation signals without increasing processing complexity or data rate.

In one embodiment, the encoded excitation signal is determined regardless of any pitch information. In particular, a preliminary excitation signal as provided by the codebook is processed to substantially eliminate at least partly representable components by the reference component at least partially related to the intermediate pitch vector. In particular, the vector components associated with the intermediate pitch vectors are removed from the preliminary excitation signal (process known as orthogonalization). The orthogonalized preliminary excitation signal is then compared with the unencoded speech sample to identify a preliminary excitation signal that best represents this particular speech sample. Then, the pitch information including the pitch filter coefficient parameter is optimized to best match the selected excitation signal, thereby optimally representing the entire speech signal.

In another embodiment, a second codebook of preliminary excitation signals, ie two excitation signals used to represent speech samples, is provided. The first excitation signal is selected as described above and the second excitation signal is selected in a similar manner, where the second preliminary excitation signal is optimized first in relation to both the intermediate pitch vector and the first selected excitation signal.

Detailed description of the invention

The present invention can be realized with a voice coder using a suitable digital signal processor such as the Motorola DSP 56000 device. The computational functions of such a DSP embodiment are shown in FIG. 1 in a block diagram.

Pitch period parameter 101 is provided to pitch filter state 102 having a pitch filter portion (determined according to the prior art). The final signal 103 has an intermediate pitch vector provided to the first multiplier 104 and to two orthogonal processing units 106 and 107 described in more detail below. The first multiplier 104 functions to generate the pitch filter output 109 by multiplying the final signal by the pitch filter coefficient 108. Pitch filter coefficient 108 selection will be described in more detail below.

The first codebook 111 includes a set of base vectors that are linearly combined to form a plurality of final excitation signals. Depending on the memory size utilized and other factors suitable for the application, the possible number of final excitation signals may be between 64 and 2048, for example, when suitable for a particular application. The problem with encoding a particular speech sample is that which female source is selected among those female sources that best represent the corresponding components of the original sound information.

According to the present invention, once the specific final signal 103 is determined, the excitation signal formulated by the first codebook 111 is provided in sequential form as a preliminary excitation source. Each preliminary excitation source is first orthogonalized 106 in relation to the final signal. For example, referring to FIG. 2, if vector A is considered to represent the final signal and vector B represents a particular preliminary excitation source, orthogonalizing the preliminary excitation signal source is represented by the reference letter B '. Results in a vector (in practice, the vector dimension space is a function of the number of samples containing the vector, which may be 40 samples or more. In addition, the preliminary excitation vector is easily orthogonalized by orthogonalizing the base vector, where Linear combination of orthogonal base vectors with each other results in an orthogonal excitation vector).

Once orthogonalized, the generated preliminary excitation source is compared 112 with an unencoded signal 113 (or a signal properly represented based thereon) to determine the relative similarity or disproportionation between the two. Then, the process is repeated for each excitation source of the first codebook 111. A determination is then made as to which preliminary excitation source most closely aligns with the unencoded signal 113.

In this particular embodiment, the gain factor 114 is also used to modify each pre-excitation signal source as can be understood by the present technology. In addition, if desired, excitation selection and gain compensations can be achieved in a nearly similar manner as can be seen with the present technology.

Once the appropriate excitation source is selected from the first codebook 11 through this process, then the orthogonalization process 106 does not need to be performed and the correct excitation signal is selected via the appropriate control mechanism 117 (116). . Subsequently, assuming a single codebook encoder, the pitch information may be derived from a selected excitation source having an excitation gain 114 and a pitch filter coefficient 108 that are optimized such that the combined excitation source is most closely aligned with the encoded signal 113. Gated together and added up. Once optimized, the pitch period parameters, pitch filter coefficients, and specific excitation and gain are known and then their proper representation is utilized as representing the original sample.

If desired, as shown in FIG. 1, an additional codebook 121 is utilized, and the second codebook again contains a plurality of basic vectors that generate preliminary excitation sources. The present technology demonstrates the use of such multiple codebooks. However, according to the present invention, once the first excitation source is selected from the first codebook 111 as described above, the preliminary excitation source from the second codebook 121 is transferred from the final signal 103 and the first codebook 111. Orthogonalized 107 in relation to the selected excitation source signal. The selection process is then continued as described above with the orthogonal preliminary excitation signal from the second codebook compared with the unencoded signal 113 to identify the most suitable one. Then, once the excitation source is selected, the pitch filter coefficients 108 and the excitation gains 114 and 120 can be optimized as described above.

Claims (9)

A speech sample encoding method comprising the steps of: A) determining a pitch period parameter for a speech sample, B) determining a plurality of encoded excitation signals of the speech sample irrespective of an arbitrary pitch filter parameter coefficient, and C) a plurality of Processing each encoded excitation signal of to provide a plurality of processed preliminary excitation signals, each of the plurality of processed preliminary excitation signals representing a pitch filter output derived at least in part as the pitch period parameter function; Providing said plurality of processed preliminary excitation signals comprising information substantially independent of and D) optimizing at least one pitch filter coefficient parameter for said speech sample. Way. 4. The method of claim 1, wherein the step of determining the encoded excitation signal processes the plurality of preliminary excitation signals to orthogonalize the plurality of preliminary excitation signals in relation to a pitch filter output that is at least partially derived as a pitch period parameter function. The voice sample encoding method further comprises. 2. The method of claim 1, wherein the step of determining the encoded excitation signal comprises: B1) processing the excitation signal to substantially eliminate a component that can be at least partially represented by a reference component that is at least partially related to the pitch period parameter; B2) determining the appropriate excitation signal for the speech sample. 4. The method of claim 3, wherein the exciting signal processing step further comprises processing the excitation signal to orthogonalize the excitation signal in relation to a pitch filter output that is at least partially derived as the pitch period parameter function. Voice sample encoding method. 4. The method of claim 3, wherein C1) substantially eliminating a reference component that is at least partially related to the pitch period parameter processing the preliminary excitation signal and a component that can at least partially represent the appropriate excitation signal determined in step C. The voice sample encoding method further comprises. 6. The method of claim 5, wherein the preliminary excitation signal processing further comprises processing the preliminary excitation signal to orthogonalize the preliminary excitation signal relating to both the reference component and the appropriate excitation signal determined in step C. A voice sample encoding method. A method of encoding a signal sample using at least two codebooks comprising information related to a preliminary excitation signal, the method comprising: A) determining a first excitation signal for the signal sample using a first codebook of the codebooks; And B) determining a second excitation signal using information of the codebook that is substantially independent of information that can be represented as the first excitation signal for a signal sample using a second codebook of the codebook; And at least partially representing said signal sample using a first and a second excitation signal. 8. The method of claim 7, wherein the signal sample further comprises a voice sample. 8. The method of claim 7, wherein determining the second excitation signal further comprises processing a preliminary excitation signal to orthogonalize the preliminary excitation signal in relation to the first excitation signal.