JP5182112B2 - Decoding device and speech coding method estimation method - Google Patents

Description

  The present invention relates to a decoding apparatus and a speech coding method estimation method.

  In recent years, various speech coding schemes targeting speech bands have been developed, and speech data encoded by the various speech coding schemes (hereinafter referred to as encoded data) is used for communication. On the other hand, in the Japanese radio law, transmission of communication waves using encoded data of a voice encoding method that has not been reported in advance is prohibited. When monitoring a communication wave for the purpose of preventing terrorism, crime, etc., it is necessary to estimate the voice encoding method of the transmitted communication wave and determine whether or not the voice encoding method has been reported.

  In order to estimate the speech coding method, there is a method in which coded data is sequentially decoded by a speech coding method decoder whose standard is disclosed, and discriminated by whether or not speech is output.

  Note that Patent Document 1 presents a method of automatically determining a speech encoding method by taking a distribution from the features of each speech encoding method with respect to encoded data and viewing the characteristics of the distribution.

JP 2007-243650 A

  A circuit for determining the output level or frequency band of audio is required to sequentially decode the encoded data with a decoder of the audio encoding method whose standard is publicly available and determine whether or not the audio is output. This complicates the configuration of the decoding apparatus. When it is determined whether or not sound is output from the output level or output frequency band of the sound, a determination error may occur due to the influence of background noise or the like. Although it is possible to determine whether or not a sound is output based on the sound of the operator, it requires a great deal of labor from the operator. Currently, there are about 20 types of speech coding schemes for speech bands whose standards are published by ITU-T (International Telecommunication Union Telecommunication Standardization Sector) or ETSI (European Telecommunications Standards Institute). If decoding is performed using all speech coding schemes whose standards are publicly available, it takes a long time to estimate the speech coding scheme.

  Further, in Patent Document 1 described above, processing for determining encoded data differs for each audio encoding method. Since it is necessary to carry out different processing for each speech encoding method, it takes time for the determination.

  An object of the present invention is to provide a decoding apparatus and a speech coding scheme estimation method that can estimate a speech coding scheme of encoded data in a short time.

  In order to achieve the above object, a decoding apparatus according to the present invention includes a data input unit that receives encoded data encoded from an unknown external audio encoding method, and a plurality of audio encoding methods that can be decoded. A decoder unit comprising: a decoder; and a frequency analysis of the encoded data received from the data input unit to estimate a frame length of an audio frame, and an audio encoding method for the encoded data based on the estimated frame length And a data processing unit that selects a decoder corresponding to the determined speech encoding method from the decoder unit, and causes the decoder to decode the encoded data.

  ADVANTAGE OF THE INVENTION According to this invention, the audio | voice coding system of coding data can be estimated in a short time.

It is a block diagram which shows the structure of the decoding apparatus of 1st Embodiment. It is a flowchart which shows the process sequence of the data processing part shown in FIG. G. 729 is a waveform diagram of a frequency spectrum of encoded data of CS-ACELP. FIG. G. 729 is a distribution diagram of a run-length histogram of encoded data of CS-ACELP (8.0 kbps). FIG. G. 729 is a distribution diagram of a run-length histogram of encoded data of CS-ACELP (6.4 kbps). FIG. G. 729 is a distribution diagram of a run-length histogram of encoded data of CS-ACELP (11.8 kbps). FIG. It is a distribution map of the run length histogram of ACELP (6.7 kbps) encoded data. It is a distribution map of the run length histogram of the coding data of LD-CELP (16 kbps). It is a distribution map of a run length histogram of encoded data of DoD MELP (54 kbps). G. 726 is a distribution diagram of a run-length histogram of encoded data of 726 ADPCM (16 kbps). FIG. G. 726 is a distribution diagram of a run-length histogram of encoded data of 726 ADPCM (24 kbps). FIG. G. 726 is a distribution diagram of a run-length histogram of encoded data of 726 ADPCM (32 kbps). FIG. G. 726 is a distribution diagram of a run-length histogram of encoded data of 726 ADPCM (40 kbps). FIG. It is a block diagram which shows the structure of the decoding apparatus of 2nd Embodiment.

  Next, the present invention will be described in detail with reference to the drawings.

  As a voice encoding method for a voice band whose standard is currently open, G.K. 729 CS-ACELP (Conjugate Structure Algebraic Code Excited Linear Prediction); 728 LD-CELP (low delay-code excited linear prediction), ACELP (Conjugate Structure Algebraic Code Excited Linear Prediction) defined by the Radio Industries Association (ARIB), MELP (United States Department of Defense defined by the US Department of Defense (DoD) Mixed Excitation Linear Prediction), G. 726 ADPCM (Adaptive Differential Pulse Code Modulation) or G722 SB-ADPCM (Sub-Band Adaptive Differential Pulse Code Modulation).

  The original audio data is encoded for each audio frame whose size is defined by each audio encoding method. Encoded data has different characteristics regarding the continuity of data values for each combination of speech encoding method and bit rate (hereinafter referred to as encoding specification). In the same coding specification, the data values change from 0 to 1, or from 1 to 0, and the bit positions in the audio frame and the run lengths representing the lengths of the same data values from the bit positions tend to be similar. It is in.

  Since the encoded data has a similar value in the speech frame generation cycle, when the encoded data is subjected to frequency analysis using a fast Fourier transform (hereinafter referred to as FFT), the multiple of the frequency at which the speech frame is generated. Wave strength increases with frequency. Using this, the frame length of the audio frame can be estimated. Since the frame length is defined for each audio encoding method and bit rate, the audio encoding method of the encoded data can be determined based on the estimated frame length.

  The decoding apparatus according to the present invention estimates the frame length of an audio frame by frequency analysis of encoded data received from the outside (hereinafter referred to as reception data), and the audio code of the received data based on the estimated frame length Determine the conversion method.

  Further, the decoding device of the present invention creates a run length histogram for each bit position of the voice frame of the received data, and compares the received data with the run length histogram created in advance for each coding specification. Is determined.

(First embodiment)
FIG. 1 is a block diagram showing the configuration of the decoding apparatus according to the first embodiment of the present invention.

  As illustrated in FIG. 1, the decoding device 1 according to the first embodiment includes a data input unit 10, a data processing unit 11, a decoder unit 12, a DA converter 13, and a database 14.

  The data input unit 10 outputs received data input from an external communication device (not shown) or a storage device such as a hard disk to the data processing unit 11.

  The data processing unit 11 performs frequency analysis on the reception data received from the data input unit 10, estimates the frame length of the speech frame, and determines the speech encoding method of the reception data based on the estimated frame length.

  In addition, the data processing unit 11 creates in advance a histogram of consecutive lengths (hereinafter referred to as reference histograms) for each bit position of an audio frame for encoded data of a predetermined size for each encoding specification. The data processing unit 11 extracts data having the same size as the reference histogram created from the received data, and creates a run length histogram for each bit position of the audio frame (hereinafter referred to as run length histogram). The data processing unit 11 compares the run length histogram and the reference histogram to determine the speech encoding method of the received data.

  Further, the data processing unit 11 compares the run length histogram with the reference histogram to estimate a bit position shift in the audio frame, discards data corresponding to the estimated bit position shift, and generates a voice synchronization shift. To compensate.

  Further, the data processing unit 11 switches the input switch 15 and the output switch 16 included in the decoder unit 12 according to the determined speech encoding method, and the received data is transmitted to the determined speech encoding method decoder included in the decoder unit 12. Output.

  The decoder unit 12 includes a plurality of decoders having different audio encoding methods that can be decoded. Examples of the decoder include a CS-ACELP decoder, an LD-CELP decoder, an ACELP decoder, a MELP decoder, an ADPCM decoder, and an SB-ADPCM decoder. Furthermore, the decoder unit 12 includes an input switch 15 and an output switch 16 for selecting a decoder to be decoded. The decoder unit 12 decodes the received data received from the data processing unit 11 with the decoder selected by the data processing unit 11 and outputs the decoded data (hereinafter referred to as decoded data) to the DA converter 13.

  The DA converter 13 converts the decoded data received from the decoder unit 12 into an analog value and outputs the analog value.

  The database 14 is a storage device for storing a reference histogram.

  The data processing unit 11 can be realized by an LSI including various logic circuits, for example.

  Next, the processing procedure of the data processing unit 11 of the decoding apparatus 1 shown in FIG. 1 will be described with reference to the flowchart of FIG.

  When the processing is started, the data processing unit 11 first waits until reception data is received from the data input unit 10 (step S1).

  When receiving the received data from the data input unit 10, the data processing unit 11 estimates the frame length of the voice frame of the received data (step S2). With the same coding specification, data values change from 0 to 1 or from 1 to 0, and bit lengths in a voice frame and run lengths indicating the lengths of the same data values after the bit positions tend to be similar. It is in. Since the encoded data has a similar value in the speech frame generation cycle, creating a frequency spectrum that replaces the encoded data with the frequency axis creates a wave strength at a frequency that is a multiple of the frequency at which the speech frame is generated. Shown to grow. A method for estimating the frame length of an audio frame using this is described in G. 729 CS-ACELP speech coding method will be described as an example. FIG. 729 is a waveform diagram of a frequency spectrum of encoded data of CS-ACELP. FIG. The horizontal axis in FIG. 3 indicates the frequency, and the vertical axis indicates the strength of the wave at each frequency. Note that the sampling frequency of the speech coding method for speech bands whose standards are currently open is 16 kHz for G722 SB-ADPCM, and 8 kHz otherwise. In the present invention, the maximum frequency extracted by FFT is 4 kHz, which is half of the minimum sampling frequency among the assumed sampling frequencies. G. In 729 CS-ACELP, since the sampling frequency is 8 kHz and the frame length of the audio frame is 80 bits, the audio frame is generated at a cycle of 100 Hz. The frequency spectrum shown in FIG. 3 shows that the intensity of the wave increases at a frequency that is a multiple of 100 Hz. That is, the frame length of the speech frame of the encoded data can be estimated by measuring the period in which the intensity of the frequency spectrum wave in which the encoded data is arranged on the frequency axis increases.

  Therefore, first, the data processing unit 11 creates a frequency spectrum in which the received data is arranged on the frequency axis using FFT.

  Next, the data processing unit 11 extracts a wave intensity peak from the frequency spectrum. A frequency spectrum assigned to the frequency axis is scanned in order from a low frequency region to a high frequency region within a preset range, and the intensity of each frequency wave is compared with preset upper and lower thresholds. A point where the wave intensity becomes maximum between the state where the wave intensity is equal to or higher than the upper limit threshold and the state where the wave intensity is equal to or lower than the lower limit threshold is extracted as a peak.

  Subsequently, the data processing unit 11 measures an interval between adjacent peaks and calculates an average value of all the peak intervals. Since the frequency spectrum is arranged on the frequency axis, the average value of the calculated peak intervals indicates the frequency. It is considered that the encoded data has a similar value at a period indicated by the reciprocal of the frequency. For this reason, it can be estimated that the voice frame is generated at the frequency. Therefore, it can be estimated that the value obtained by dividing the frequency by the sampling frequency is the frame length of the audio frame. As shown in FIG. In the frequency spectrum of the encoded data of 729 CS-ACELP, since the average value of the peak interval is 100 Hz and the sampling frequency is 8 kHz, the frame length of the voice frame is estimated to be 80 frames. Since the frame length of the audio frame is defined according to the encoding specification, the encoding specification can be estimated from the estimated frame length. In step S2, since the sampling frequency of the received data is not known, the frame length of the audio frame is calculated on the assumption that the sampling frequency is 8 kHz. The calculated frame length is defined as a specified frame length. When the speech encoding method is G722 SB-ADPCM, the sampling frequency is 16 kHz, and the specified frame length is half of the actual frame length.

  When the prescribed frame length is estimated, the data processing unit 11 compares the estimated prescribed frame length with the prescribed frame lengths of all the coding specifications for which the standards are disclosed (step S3). In step S2, the data processing unit 11 calculates the specified frame length on the assumption that the sampling frequency of the received data is 8 kHz. For this reason, when compared with a speech coding method having a sampling frequency of 16 kHz, half the actual frame length of the coding specification is set as the specified frame length. When the estimated specified frame length matches the specified frame length of the encoding specification for which the standard is published, the data processing unit 11 picks up the encoding specification as a candidate for the encoding specification of the received data.

  If none of the encoding specifications for which the standard is published is picked up as a candidate for the encoding specification of the received data, the data processing unit 11 determines that the speech encoding method of the received data is unknown, and proceeds to step S9. To do. When the process proceeds to step S9, the data processing unit 11 notifies the user that the speech encoding method is unknown (step S9), and ends the process.

  On the other hand, when one or more encoding specifications are picked up as candidates for the encoding specification of the received data, the data processing unit 11 proceeds to step S4 and creates a continuous length histogram of the received data (step S4).

  FIG. 729 is a distribution diagram of a run-length histogram of encoded data of CS-ACELP (8.0 kbps). FIG. The left diagram of FIG. 4 shows a run length histogram of 0, and the right diagram shows a run length histogram of 1. The Y axis in FIG. 4 indicates the bit position in the audio frame, the X axis indicates the run length number, and the Z axis indicates the run length frequency in a log scale. The maximum value of the Y axis is the specified frame length estimated in step S2.

  An example of the procedure for creating the distribution chart of the run length histogram shown in FIG. 4 is shown below. The data processing unit 11 scans the received data from the beginning, and when the data value changes from 1 to 0, the bit position in the audio frame in which the data value has changed is set as a value on the Y axis, and the data value 0 from the bit position. Are plotted on a run-length histogram of 0, with the length of continuous as a value on the X-axis. Similarly, when the data value changes from 0 to 1, the data processing unit 11 sets the bit position in the audio frame where the data value has changed to a value on the Y axis, and the length from which the data value 1 continues from the bit position. Is plotted on the run length histogram of 1 as the value on the X-axis. The continuous length of data value 0 is a run length of 0, and the continuous length of data value 1 is a run length of 1.

  When the data processing unit 11 creates the run length histogram of the received data, it compares the created run length histogram with reference histograms of all the coding specifications picked up as candidates for the coding specification of the received data in step S3 (step S3). S5).

  FIG. 729 is a distribution diagram of a run-length histogram of encoded data of CS-ACELP (6.4 kbps). FIG. FIG. 729 is a distribution diagram of a run-length histogram of encoded data of CS-ACELP (11.8 kbps). FIG. FIG. 7 is a distribution diagram of a run length histogram of ACELP (6.7 kbps) encoded data. FIG. 8 is a distribution diagram of a run length histogram of LD-CELP (16 kbps) encoded data. FIG. 9 is a distribution diagram of a run length histogram of DoD MELP (54 kbps) encoded data. FIG. 726 is a distribution diagram of a run-length histogram of encoded data of 726 ADPCM (16 kbps). FIG. FIG. 726 is a distribution diagram of a run-length histogram of encoded data of 726 ADPCM (24 kbps). FIG. FIG. 726 is a distribution diagram of a run-length histogram of encoded data of 726 ADPCM (32 kbps). FIG. FIG. 726 is a distribution map of a run length histogram of encoded data of 726 ADPCM (40 kbps); As shown in FIGS. 4 to 13, the run length histogram has different characteristics depending on the encoding specification. In the same coding specification, the data position changes from 0 to 1 or from 1 to 0, and the bit position in the audio frame and the run length at the bit position tend to be similar. Therefore, by comparing the run length histogram created in step S4 with a reference histogram created in advance for each coding specification for which the standard is published, and obtaining the correlation thereof, the speech coding scheme of the received data is determined. Can be estimated.

The 0 run length histogram created in step S4 is f ₀ (x, y), and the 1 run length histogram is f ₁ (x, y). In addition, a reference histogram of 0 created in advance for each coding specification for which a standard is disclosed is g ₀ (x, y), and a reference histogram of ₁ is g ₁ (x, y). Here, x is the number of run lengths at bit positions in each audio frame. If the maximum value of the run length number is M, the value is in the range of 1 ≦ x ≦ M. Further, y is a bit position in the audio frame, and is a value in the range of 0 ≦ y ≦ (N−1), where N is the specified frame length of the audio frame.

  Further, the decoding device 1 is not necessarily synchronized with an encoding device (not shown). For this reason, the decoding apparatus 1 does not necessarily receive the encoded data encoded by the encoding apparatus from the beginning. Therefore, the correlation between the run length histogram created in step S4 and the reference histogram must be obtained in consideration of the bit position shift in the audio frame.

  Therefore, the data processing unit 11 obtains a correlation value with the reference histogram by shifting the bit position in the speech frame of the run length histogram by a value k. Note that k is a value within the range of the bit position in the audio frame, and does not exceed the specified frame length of the audio frame.

  As shown in the equation (1), the correlation value is calculated as the sum of the absolute difference value of the 0 run length and the absolute difference value of the run length of 1 for each bit position in the speech frame of the run length histogram and the reference histogram. The

  In the range of 0 ≦ k ≦ (N−1), it can be estimated that k at which the correlation value calculated by the equation (1) is the minimum is a bit position shift in the audio frame.

  The data processing unit 11 calculates correlation values for the reference histograms of all the encoding specifications picked up as candidates for the encoding specifications of the received data in step S3. The data processing unit 11 estimates that the speech coding method of the reference histogram that minimizes the correlation value is the speech coding method of the received data. If there is only one type of encoding specification picked up as a candidate for the encoding specification of the received data in step S3, it is estimated that the speech encoding method of the encoding specification is the speech encoding method of the received data.

  Subsequently, the data processing unit 11 determines whether or not the correlation value calculated in step S5 is equal to or less than a preset threshold value (step S6). If the correlation value calculated in step S5 is not less than or equal to a preset threshold value, the data processing unit 11 determines that the speech encoding method of the received data is unknown, and proceeds to step S9. When the process proceeds to step S9, the data processing unit 11 notifies the user that the speech encoding method is unknown (step S9), and ends the process.

  On the other hand, if the correlation value calculated in step S5 is less than or equal to a preset threshold value, the data processing unit 11 determines that the received data has been encoded by the speech encoding method of the encoding specification estimated in step S5. The process proceeds to S7.

  In step S7, the data processing unit 11 discards data corresponding to the bit position shift in the audio frame estimated in step S6 from the beginning of the received data, and synchronizes the audio frame (step S7).

  Subsequently, the input switch 15 and the output switch 16 included in the decoder unit 12 are switched to select the estimated speech coding decoder (step S8), and the process is terminated.

  The decoding apparatus according to the first embodiment proceeds to step S4 after estimating the frame length in step S2, and limits the speech coding scheme based on the continuous length histogram of the received data. If the speech encoding scheme of the received data can be limited when the frame length is estimated in step S2, steps S4 to S7 may be omitted, and a speech encoding scheme decoder estimated from the frame length may be selected. In this case, the data processing unit 11 cannot perform synchronization deviation compensation, and the decoder unit 12 needs to perform synchronization deviation compensation. However, since steps S4 to S7 can be omitted, the speech coding method for received data at a higher speed. Can be estimated.

  The decoding apparatus according to the first embodiment estimates the frame length of an audio frame by performing frequency analysis on received data. If the estimated frame length matches the frame length defined in only one type of coding specification for which the standard is published, the speech coding scheme of the coding specification is the received data It can be estimated that this is a speech coding method. Also, if the estimated frame length does not match the frame length defined in all coding specifications for which the standard is published, it can be estimated that the received data is encoded by an unknown speech encoding method .

  Further, the decoding apparatus according to the first embodiment creates a continuous length histogram of received data, and compares the received data with a reference histogram created in advance for each coding specification whose standard is published. Is estimated. If the speech encoding method cannot be limited based on the frame length of the speech frame of the received data, the speech encoding method can be limited to one type by comparing the run length histogram and the reference histogram. Also, the accuracy of estimation of the speech coding scheme of received data can be increased by comparing the run length histogram and the reference histogram.

  In addition, the decoding apparatus according to the first embodiment obtains the correlation with the reference histogram in consideration of the bit position shift in the audio frame of the run length histogram. For this reason, even when the decoding device and the encoding device are not synchronized, the decoding device can estimate the voice encoding method of the received data. Since the decoding device supplies the decoder unit with received data in which data corresponding to the bit position shift in the audio frame is discarded, it is not necessary to synchronize the audio frame in the decoder unit, and the time until the decoding process is performed. Can be shortened.

  As a result, it is possible to estimate the speech encoding method of the received data in a short time.

(Second Embodiment)
The processing of the data processing unit of the first embodiment may be realized by a program.

  FIG. 14 is a block diagram showing the configuration of the decoding apparatus according to the second embodiment.

  As shown in FIG. 14, the decoding device 1 according to the second embodiment includes a CPU 20, a main storage device 21, an auxiliary storage device 22, a recording medium interface device 23, a recording medium 24, and a data input unit 10. A database 14, a decoder unit 12, and a DA converter 13, which are connected via an internal bus 30.

  The recording medium 24 records a program for realizing the function of the data processing unit of the first embodiment. The program recorded on the recording medium 24 is read into the main storage device 21 by the CPU 20 via the storage medium interface device 23. The CPU 20 executes processing according to the program read into the main storage device 21. The recording medium 24 may be a magnetic disk, a semiconductor memory, an optical disk, or other recording medium.

DESCRIPTION OF SYMBOLS 1 Decoding apparatus 10 Data input part 11 Data processing part 12 Decoder part 13 DA converter 14 Database 15 Input switch 16 Output switch 20 CPU
21 Main storage device 22 Auxiliary storage device 23 Recording medium interface unit 24 Recording medium

Claims (6)

A data input unit for receiving encoded data encoded by an unknown speech encoding method from the outside;
A decoder unit comprising a plurality of decoders having different decodable audio encoding methods;
A frame length of a speech frame is estimated by frequency analysis of the encoded data received from the data input unit, a speech encoding method of the encoded data is determined based on the estimated frame length, and the decoder unit A data processor that selects a decoder corresponding to the determined speech encoding method and causes the decoder to decode the encoded data;
A decoding device. The data processing unit
A histogram of the run length for each bit position of the audio frame of the encoded data is created, and for each bit position of the audio frame created in advance for each combination of the audio encoding method and the bit rate for which the standard is disclosed The decoding apparatus according to claim 1, wherein a speech encoding method of the encoded data is determined by comparison with a run length histogram. The data processing unit
A continuous length histogram for each bit position of the audio frame of the encoded data and a sequence for each bit position of the audio frame created in advance for each combination of the audio encoding method and bit rate for which the standard is disclosed. 3. The decoding according to claim 2, wherein a deviation of a bit position in a voice frame is estimated by comparing length histograms, and the encoded data in which data corresponding to the estimated deviation of the bit position is discarded is supplied to the decoder unit. apparatus. A speech encoding method estimation method for a decoding device including a plurality of decoders having different speech encoding methods that can be decoded from each other,
Estimate the frame length of the voice frame by frequency analysis of the received encoded data,
Determining a speech encoding method of the encoded data based on the estimated frame length;
A speech encoding method estimation method for selecting a decoder corresponding to the determined speech encoding method and causing the decoder to decode the encoded data.   A histogram of the run length for each bit position of the audio frame of the encoded data is created, and for each bit position of the audio frame created in advance for each combination of the audio encoding method and the bit rate for which the standard is disclosed 5. The speech coding method estimation method according to claim 4, wherein a speech coding method of the coded data is determined by comparing with a run length histogram. A continuous length histogram for each bit position of the audio frame of the encoded data and a sequence for each bit position of the audio frame created in advance for each combination of the audio encoding method and bit rate for which the standard is disclosed. By comparing the long histograms, we estimate the bit position shift in the audio frame,
6. The speech encoding method estimation method according to claim 5, wherein the encoded data in which data corresponding to the estimated bit position shift is discarded is supplied to the decoder.

Images