JP2011170259A - Voice encoding device, method and program, and code book data classification device, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice encoding device enabling processing required for code search to be reduced while maintaining compatibility with existing various standards and sound quality. <P>SOLUTION: The voice encoding device utilizing a fixed code book, has: a signal characteristic analysis means which analyzes a characteristic of the voice signal to be encoded; and a candidate code group changing means for changing, according to the analysis result, a code group which is used for searching an optimal code to be included in encoding information among codes in the fixed code book. Since a code used for searching the optimal code among the codes of the fixed code book is limited according to the characteristic of the voice signal to be encoded, the encoding speed is increased. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a speech encoding apparatus, method and program, and codebook data classification and aggregation apparatus, method and program, and can be applied to, for example, a mobile phone terminal.

  In telephone communications (particularly mobile telephone communications) that are voice communications, voice coding techniques are used to reduce the amount of information to be transmitted. For speech coding, a technique called CELP (Code Excluded Linear Prediction) type is widely used and has been put into practical use as many standards. CELP speech coding has a quantization table called a code book, and realizes coding by selecting an optimum code from this table. In general, in CELP speech coding, an optimal code is searched using all the codes in the code book as candidates, so that a lot of processing is required for code search (code selection). For this reason, various speed-up methods have been devised. For example, there is a standard method in which high-speed processing as described in Non-Patent Document 1 is considered in advance.

ITU-T G. 729A

  However, in addition to the technique described in Non-Patent Document 1, there are many standard systems using CELP speech coding, and the technique described in Non-Patent Document 1 cannot be applied to such a standard system.

  Therefore, a new technique that can be applied to many schemes and standards and that can execute speech encoding at high speed is desired.

  According to a first aspect of the present invention, in a speech encoding apparatus using a fixed codebook, (1) a signal characteristic analyzing means for analyzing characteristics of a speech signal to be encoded, and (2) an analysis result of the signal characteristic analyzing means And a candidate code group switching means for switching a code group used for searching for an optimum code to be included in the encoded information among the codes in the fixed codebook.

  According to a second aspect of the present invention, in the speech encoding method using a fixed codebook, (1) the signal characteristic analyzing means analyzes the characteristics of the speech signal to be encoded, and (2) the candidate code group switching means is According to an analysis result of the signal characteristic analysis means, a code group used for searching for an optimum code to be included in encoded information is switched among codes in the fixed codebook.

  According to a third aspect of the present invention, there is provided a speech encoding program for executing speech encoding using a fixed codebook, wherein (1) signal characteristic analysis means for analyzing characteristics of a speech signal to be encoded And (2) functioning as a candidate code group switching means for switching a code group used for searching for an optimum code to be included in encoded information among codes in the fixed codebook according to the analysis result of the signal characteristic analysis means. Features.

  The code book data classification and aggregation device according to the fourth aspect of the present invention comprises: (1) signal characteristic analysis means for analyzing characteristics of audio signals; and (2) optimal code search using all codes in a fixed codebook as candidates. Candidate code group to be used for speech coding at the time of the analysis result for each analysis result among the codes of the fixed codebook, by summing up the frequency information as codes for each analysis result of the signal characteristic analysis means Codebook data classification and aggregation means for obtaining the above information.

  In the code book data classification and aggregation method of the fifth aspect of the present invention, (1) the signal characteristic analysis means analyzes the characteristics of the audio signal, and (2) the code book data classification and aggregation means analyzes all the codes in the fixed code book. The frequency information that has become the optimal code in the search for the optimal code as a candidate is aggregated for each analysis result of the signal characteristic analysis means, and for each analysis result of the code of the fixed codebook, Information on candidate code groups used in speech coding is obtained.

  A codebook data classification and aggregation program according to a sixth aspect of the present invention includes: (1) a signal characteristic analysis unit that analyzes the characteristics of an audio signal; and (2) an optimal code that uses all codes in a fixed codebook as candidates. The frequency information that has become the optimum code in the search is aggregated for each analysis result of the signal characteristic analysis means, and is used for speech encoding at the time of the analysis result for each analysis result in the code of the fixed codebook. It is made to function as a code book data classification totaling means for obtaining information on candidate code groups.

  According to the present invention, the codes used for searching for the optimum code among the codes of the fixed codebook are limited according to the characteristics of the audio signal to be encoded. The processing required for chord search can be reduced while maintaining sound quality.

It is a block diagram which shows the functional structure of the audio | voice coding apparatus which concerns on 1st Embodiment. It is explanatory drawing of voiced sound candidate information and unvoiced sound candidate information in 1st Embodiment. It is a block diagram which shows the functional structure of the audio | voice coding apparatus which concerns on 2nd Embodiment.

(A) First Embodiment Hereinafter, a first embodiment of a speech encoding apparatus, method, and program according to the present invention will be described in detail with reference to the drawings. The first embodiment does not include a codebook data classification and aggregation device, method, and program.

  FIG. 1 is a block diagram showing a functional configuration of the speech encoding apparatus according to the first embodiment. FIG. 1 shows components for searching (selecting) an optimal code. When the device equipped with the speech encoding device is a mobile phone terminal or the like, the speech encoding device is often realized by a CPU and a speech encoding program executed by the CPU. However, it can be functionally represented in FIG.

  In FIG. 1, a speech encoding apparatus 1 according to the first embodiment includes a codebook unit 2, a gain variable unit 3, a synthesis filter unit 4, a subtraction unit 5, an auditory weighting unit 6, an optimal code evaluation unit 7, and a signal. A characteristic analysis unit 8 is included. The code book unit 2 includes a code book 10 and a code extraction unit 11.

  The speech coding apparatus 1 according to the first embodiment is characterized by the code book unit 2 and the signal characteristic analysis unit 8, and other configurations are shown in FIG. 1 as long as the configuration conforms to the CELP speech coding scheme. The configuration is not limited.

  The speech encoding apparatus 1 according to the first embodiment performs, for example, encoding for each frame of speech signals (speech data) for a predetermined time, and is optimal for the speech signal of the frame for each frame. Search for code.

  The code book unit 2 has a code book 10 in which a predetermined number of predetermined codes are stored in a fixed manner, and the code extraction unit 11 stores an optimum code from the code book 10 as will be described later. Code candidates applied to the search are sequentially extracted.

  The code extracted from the code book unit 10 is given to the synthesis filter unit 4 after the gain is adjusted by the gain variable unit 3. The synthesis filter unit 4 is also given a vocal tract coefficient obtained by quantizing a vocal tract coefficient obtained by a vocal tract coefficient analysis unit (not shown) from a speech signal to be encoded by a vocal tract coefficient quantization unit (not shown). The synthesis filter unit 4 performs synthesis processing from a given code and vocal tract coefficient, and obtains a speech signal similar to that obtained by local decoding by applying a code candidate (hereinafter, locally decoded speech signal). The subtracting unit 5 obtains the speech signal to be encoded and the locally decoded speech signal, and the error signal (difference signal) is weighted according to the auditory characteristics by the auditory weighting unit 6, and then the optimum code The power of the weighted error signal (square error) is calculated by the evaluation unit 7. The optimal code evaluation unit 7 determines a code candidate having the minimum square error as the optimal code from the square errors for the respective code candidates.

  An audio signal to be encoded is input to the signal characteristic analysis unit 8, and the signal characteristic analysis unit 8 analyzes the characteristic and gives a parameter representing the analysis result of the characteristic to the code extraction unit 11. In the following description, it is assumed that the parameter output from the signal characteristic analyzer 8 indicates whether the parameter is voiced sound or unvoiced sound. However, the analysis characteristic by the signal characteristic analysis part 8 is not limited to this. For example, it is possible to determine (analyze) sound / silence. Further, for example, it may be an analysis of whether the pitch is in a short range, medium range, or long range.

  The code extraction unit 11 includes code information (voiced sound candidate information) 11a that can be a candidate for voiced sound and code information (unvoiced sound candidate information) 11b that can be a candidate for unvoiced sound.

  FIG. 2 is an explanatory diagram of the voiced sound candidate information 11a and the unvoiced sound candidate information 11b. The code book 10 stores a predetermined number (eg, 1024 types) of codes. When the speech signal to be encoded is a voiced sound, all the codes in the codebook 10 are not used for searching for the optimum code, but are described in the voiced sound candidate information 11a among the codes stored in the codebook 10. A part of the codes (for example, 512 types) are used for searching for the optimum code. Similarly, when the speech signal to be encoded is an unvoiced sound, all the codes in the code book 10 are not used for searching for the optimum code, but are described in the unvoiced sound candidate information 11b among the codes stored in the code book 10. A part of the codes (for example, 512 types) are used for searching for the optimum code.

  As the voiced sound candidate information 11a and the unvoiced sound candidate information 11b, for example, a sequence of bits indicating whether or not to use the codes is applied in the order of the codes stored in the code book 10.

  In the example of FIG. 2, the code stored first in the code book 10 is not a candidate for voiced sound but a candidate for unvoiced sound. The second code stored in the code book 10 is a candidate for a voiced sound and not a candidate for an unvoiced sound. The fourth code stored in the code book 10 is a candidate for both voiced and unvoiced sounds. The sixth code stored in the code book 10 is neither a voiced sound nor an unvoiced sound.

  When the audio signal (frame) to be encoded is a voiced sound, the code extraction unit 11 sequentially extracts and outputs codes that can be candidates from the codebook 10 in accordance with the voiced sound candidate information 11a. When the audio signal (frame) is an unvoiced sound, codes that can be candidates are sequentially extracted from the code book 10 and output according to the unvoiced sound candidate information 11b.

  A method for forming the voiced sound candidate information 11a and the unvoiced sound candidate information 11b stored in the code extraction unit 11 will be clarified in the description of the second embodiment to be described later. The code described in the voiced sound candidate information 11a is a code that is often the optimal code in the voiced sound frame, and the code described in the unvoiced sound candidate information 11b is the optimal code in the unvoiced sound frame. This is a code that has often become.

  Note that since the information of the optimal code is given to the speech decoding apparatus and decoding is performed based on the information, the speech decoding apparatus is not provided even if a configuration characteristic to the speech encoding apparatus is incorporated as in the first embodiment. It can operate in the configuration.

  According to the first embodiment, the codes used for the search for the optimum code among the codes of the code book are limited according to the characteristics of the audio signal (frame) to be encoded. Processing (processing amount, processing time, etc.) required for chord search (code selection) can be reduced while maintaining compatibility with standardization and sound quality.

(B) Second Embodiment Next, a second embodiment of the speech encoding apparatus, method and program, and codebook data classification and aggregation apparatus, method and program according to the present invention will be described in detail with reference to the drawings. To do. Here, the speech encoding apparatus according to the second embodiment includes the code book data classification and aggregation apparatus according to the second embodiment.

  FIG. 3 is a block diagram showing a functional configuration of the speech coding apparatus according to the second embodiment. The same reference numerals are used for the same and corresponding parts as in FIG. 1 according to the first embodiment (however, (Excluding the reference numeral).

  In FIG. 3, a speech encoding apparatus 1A according to the second embodiment includes a codebook unit 2, a gain variable unit 3, a synthesis filter unit 4, a subtraction unit 5, and an auditory weighting unit 6 similar to those in the first embodiment. In addition to the optimum code evaluation unit 7 and the signal characteristic analysis unit 8, a code book data classification totaling unit 20 and a classification totaling activation switch 21 are provided.

  The classification / aggregation activation switch 21 is a switch that activates classification of all codes in the codebook 10 as belonging to the voiced sound candidate information 11a or the unvoiced sound candidate information 11b. A description will be given of an example in which the speech encoding apparatus 1A of the second embodiment is mounted on a mobile phone terminal. In the mobile phone terminal store, various settings are made when the mobile phone terminal is sold. As one setting item at that time, setting items for voiced sound candidate information and unvoiced sound candidate information are provided. By operating the classification / aggregation activation switch 21 provided inside the terminal, such as in the space of the battery compartment, the setting operation of the voiced sound candidate information and the unvoiced sound candidate information, in other words, all the codes in the code book 10 are included. The operation of classifying whether it belongs to the voice sound candidate information 11a or the unvoiced sound candidate information 11b is started. Also, the user of the mobile phone terminal is allowed to speak for a predetermined time, for example, toward the mobile phone terminal.

  In such a classification / aggregation mode, the code extraction unit 11 sequentially outputs all the codes in the code book 10 as candidate codes, and determines an optimum code.

  Such an optimal code for each frame to be encoded is given to the codebook data classification and counting unit 20, and the characteristic analysis result of the frame is also given to the codebook data classification and counting unit 20. The code book data classification and counting unit 20 counts the number of times that all the codes of the code book 10 have become the optimum codes for the voiced sound frames, and stores all the codes of the code book 10 for the unvoiced sound frames. Each tabulates the number of times that the code is optimal.

  For example, when the classification / aggregation mode ends in a predetermined time, the codebook data classification / aggregation unit 20 writes the information of a predetermined number of codes having a larger number of times of being the optimum code for the voiced sound frame into the voiced sound candidate information 11a. Then, the information of a predetermined number of codes having the larger number of times of becoming the optimum code for the unvoiced sound frame is written in the unvoiced sound candidate information 11b. In the subsequent encoding, the candidate code extracted from the code book is switched according to the characteristics of the audio signal (frame) to be encoded.

  According to the second embodiment, the same effects as those of the first embodiment can be obtained, and the voiced sound candidate information 11a and the unvoiced sound candidate information 11b according to the user can be prepared. It is possible to achieve an effect that it is possible to perform appropriate encoding according to the voice.

  In the second embodiment, the codebook data classification / aggregation unit 20 provided in the speech encoding device 1A creates the voiced sound candidate information 11a and the unvoiced sound candidate information 11b. In the case of the first embodiment, for example, a codebook data classification and aggregation device (for example, a computer) provided by a manufacturer or the like creates voiced sound candidate information 11a and unvoiced sound candidate information 11b (the creation method is the second implementation). This is the same as the embodiment), and is written in the code extraction unit 11.

(C) Other Embodiments In the second embodiment, the voiced sound candidate information 11a has been shown to define the voiced sound candidate information 11a and the unvoiced sound candidate information 11b with the frequency at the end of the classification and counting mode for a predetermined time. And the determination method of the unvoiced sound candidate information 11b is not limited to this as long as it can be determined from the one where the number of times that the optimum code is obtained is large. For example, for a voiced sound frame, a code that has reached a predetermined number of times as an optimal code is written in the voiced sound candidate information 11a, and when the number of written codes reaches a predetermined number, The writing may be terminated, and the unvoiced sound candidate information 11b may be formed similarly.

  In the above description, it has been described that the analysis characteristic by the signal characteristic analysis unit 8 is not limited to voiced sound or unvoiced sound, but may be other characteristics. The analysis characteristic by the signal characteristic analysis unit 8 may be a combination of a plurality of characteristics. For example, the analysis result may be a voiced sound with a short pitch, a voiced sound with a long pitch, or an unvoiced sound, and the candidate codes may be switched according to these three types of results.

  Further, two thresholds for determining the unvoiced sound may be prepared, and the number of code candidates may be changed depending on whether the unvoiced sound level is high or the unvoiced sound level is low (close to voiced sound).

  Furthermore, the operator may be able to select the number of candidates to be limited. For example, the operator may be able to select whether to limit to half the number of codes in the code book or to limit to 1/4 of the number of codes in the code book. That is, the operator may select whether the quality is important or the processing time is important.

  The present invention can be widely applied to a speech encoding apparatus including a code book in which codes are fixedly stored. For example, if the speech coding apparatus includes an adaptive codebook and a fixed codebook, the technical idea of the present invention can be applied to the fixed codebook. In addition, for example, if the speech coding apparatus includes two or more types of codebooks, the technical idea of the present invention can be applied to all fixed codebooks, and some fixed codebooks. In contrast, the technical idea of the present invention can be applied. When candidates are limited for a plurality of fixed codebooks, the characteristics for switching candidate groups may differ depending on the fixed codebook.

  DESCRIPTION OF SYMBOLS 1, 1A ... Speech coding apparatus, 2 ... Code book part, 7 ... Optimum code evaluation part, 8 ... Signal characteristic analysis part, 10 ... Code book, 11 ... Code extraction part, 20 ... Code book data classification totaling part, 21 … Category aggregation start switch.

Claims (7)

In a speech encoding apparatus using a fixed codebook,
Signal characteristic analysis means for analyzing the characteristics of the audio signal to be encoded;
A speech code comprising: candidate code group switching means for switching a code group used for searching for an optimum code to be included in encoding information among codes in the fixed codebook according to an analysis result of the signal characteristic analysis means. Device.   Candidate code group that aggregates the frequency information that has become the optimum code in the search of the optimum code with all the codes in the fixed codebook as candidates, and is switched by the candidate code group switching means for each analysis result of the signal characteristic analysis means The speech coding apparatus according to claim 1, further comprising code book data classification and aggregation means for obtaining the information. In a speech coding method using a fixed codebook,
The signal characteristic analysis means analyzes the characteristics of the audio signal to be encoded,
Speech coding characterized in that the candidate code group switching means switches the code group used for searching for the optimum code to be included in the coding information, among the codes in the fixed codebook, according to the analysis result of the signal characteristic analysis means Method. A speech encoding program for executing speech encoding using a fixed codebook,
Computer
Signal characteristic analysis means for analyzing the characteristics of the audio signal to be encoded;
According to an analysis result of the signal characteristic analysis means, the code function is functioned as a candidate code group switching means for switching a code group used for searching for an optimum code to be included in encoding information among codes in the fixed codebook. Speech coding program. A signal characteristic analysis means for analyzing the characteristics of the audio signal;
The frequency information that became the optimum code in the search of the optimum code with all the codes in the fixed codebook as candidates is tabulated for each analysis result of the signal characteristic analysis means, and among the codes of the fixed codebook, for each analysis result And a codebook data classification and aggregation device for obtaining information of candidate code groups used in speech encoding at the time of the analysis result. The signal characteristic analysis means analyzes the characteristics of the audio signal,
The code book data classification and counting means totals the frequency information that has become the optimal code in the search for the optimal code with all the codes in the fixed code book as candidates, for each analysis result of the signal characteristic analyzing means, and the fixed code book A codebook data classification and aggregation method characterized in that, for each analysis result, information on candidate code groups to be used for speech encoding at the time of the analysis result is obtained. Computer
A signal characteristic analysis means for analyzing the characteristics of the audio signal;
The frequency information that became the optimum code in the search of the optimum code with all the codes in the fixed codebook as candidates is tabulated for each analysis result of the signal characteristic analysis means, and among the codes of the fixed codebook, for each analysis result And a codebook data classification / aggregation program that functions as a codebook data classification / aggregation means for obtaining information on candidate code groups to be used in speech encoding at the time of the analysis result.

Images