JPS6253093B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to an audio reproduction device that reproduces an original audio signal waveform from digital data including data obtained by encoding an audio signal waveform using an adaptive delta modulation method. As is well known, the delta modulation method, which is one of the audio signal waveform encoding methods, uses a sufficiently large sampling frequency to increase the correlation between adjacent sample values so that the difference in amplitude falls within a certain range. This is a method in which the magnitude relationship of the amplitudes is then expressed and encoded using one bit, ``1'' or ``0''. It is known that this delta modulation method is accompanied by two types of distortion in principle. One of them is called slope overload, and as shown in Figure 1a, the output of the local D/A converter in the delta modulation encoder responds to a steep change in the input audio signal waveform 1. This is distortion caused by the inability of waveform 2 to follow. 3 indicates encoded data. The other type is called granular distortion. As shown in Figure 1b, when the input audio signal waveform 1 does not change much over time, the output waveform of the local D/A converter 2 operates increase/decrease,
This is due to the fact that the encoded data 3 alternately repeats "1" and "0". An adaptive delta modulation method that adaptively changes the quantization width depending on the amplitude of the audio signal was devised as a method for reducing these two types of distortion. In other words, if the data that encodes the audio signal waveform is continuously "1" or "0", slope overload has occurred, so the quantization width is increased, and when the data is constantly changing, the quantization width is increased. Since distortion occurs, the quantization width is conversely reduced. Figures 2a and b show CVSD (Continuous Variable Slope Delta), which is a type of adaptive delta modulation method.
2 shows an example of a configuration of an encoder side and a decoder side in an audio recording and reproducing apparatus using a modulation method. On the encoder side, the input audio signal 11 is passed through a low-pass filter 12 and subtracted by a comparator 13, and then retimed by a clock CLK in a D-flip-flop 14 to become encoded data. In the basic delta modulation method, the comparison level of the comparator 13 is given by the output of the estimate filter 15 (corresponding to a local D/A converter), and its step, that is, the quantization width, is constant. The width of the change is continuously and adaptively controlled. That is, if the waveform of the input audio signal 11 changes suddenly and the encoded data has 3 or more bits of "1" or "0" continuously,
This is detected by the shift register 16 and the EX-OR circuit 17, and the output of the syllabi filter 18 is integrated, so that the gain of the multiplier 19 provided on the output side of the estimate filter 15 increases, and the quantization is performed. The width increases sequentially for each bit. This suppresses slope overload. On the other hand, when the waveform of the input audio signal 11 does not change much over time, the filter 1
As the output of 8 decreases, the gain of the multiplier 19 decreases, and the quantization width gradually decreases.
Granular distortion can be suppressed. On the decoder side, Estmate filter 21,
A configuration consisting of a multiplier 22, a shift register 23, an EX-OR circuit 24, a syllabi filter 25, and a low-pass filter 26 performs an operation opposite to that of the encoder, and a reproduced audio signal 27 is obtained. However, even with adaptive delta modulation methods such as the CVSD method, there are still problems that need to be improved in terms of the clarity of reproduced speech. This problem is explained in the third
This will be explained using FIG. Figure 3a shows the envelope waveform of the audio signal when the Japanese word "Oyasuminojikan" is spoken at a normal speed, and Figure 3b shows the sound/silence discrimination signal corresponding to this audio signal. ing. In FIG. 3b, H represents a sound section and L represents a silent section. According to this, for example, the silent interval between "OYA" and "SUMI" is approximately 120 ms. Generally, in normal conversation, there are quite a few silent periods of around 100 ms, including between words and syllables. Looking at it macroscopically in relation to Figure 3, which shows the change in quantization width in the CVSD method,
Since the amplitude of "OYA" is large, the quantization width becomes large, and in the silent section between "OYA" and "SUMI", the quantization width becomes smaller due to the time constant of the Silavik filter. Normally, the time constant of the Silavik filter is selected to be about 4ms with respect to a 16kHz clock, so no matter how large the quantization width is in a sound section, in a silent section where the time width is mostly around 100ms, the quantization width is The distortion width settles to the minimum value, and granular distortion is suppressed. Therefore, a sound section following a silent section, for example "SUMI"
At the beginning of SUMI, encoding is performed with the minimum quantization width, and from a microscopic perspective, flop overload occurs in most cases at the beginning of SUMI. As a general characteristic of speech signals, the starting point of each syllable is an important part related to phonology and intelligibility.
For example, if we look at the envelope waveform of a plosive in a monosyllable, we can see that it is 10 to 100 meters from the starting point, as shown in Figure 4.
The period of s includes plosive part A, buzz B, aspiratory part C, and exit part D, etc., and if these parts are not encoded without distortion, the sound will sound like a different sound and the phonology will be distorted. The result is that the In this way, in conventional adaptive delta modulation methods such as CVSD, quantization is performed in response to changes in the audio signal waveform, especially at the beginning of a voiced period after a relatively long silent period and a state in which the quantization width has reached an almost minimum value. Due to the poor followability of changes in width, slope overload is likely to occur, which causes loss of clarity of reproduced audio. An object of the present invention is to improve intelligibility by compensating for slope overload occurring at the beginning of a sound section when reproducing an original audio signal waveform from digital data including data encoded by adaptive delta modulation. The purpose of the present invention is to provide an audio playback device that can perform the following tasks. The present invention detects a silent section of digital data read from a memory, and instructs a decoder to make a quantization width larger than a minimum quantization width instead of a silence pattern during part or all of the silent section. It is characterized by supplying specific digital data and maintaining the output of the decoder at a silent level during that period. In this way, the quantization width is forcibly increased rapidly in the decoder at the beginning of a voiced period, thereby compensating for the slope overload at the beginning of the voiced period, which has a large effect on intelligibility, during playback. Can be done. Furthermore, since the output of the decoder is maintained at a silent level during the period when this specific digital data is supplied to the decoder, noise caused by this digital data does not occur in the reproduced audio signal waveform. Embodiments of the present invention will be described below with reference to the drawings. FIG. 5 shows the configuration of an audio reproduction device according to an embodiment of the present invention. In the figure, a memory 51 stores digital data including data encoded by the adaptive delta modulation method of an audio signal waveform, for example, data encoded by the encoder shown in FIG. 2a. The contents of this memory 51 are read out via a controller 52, and the original audio signal waveform is reproduced via a decoder 53. The decoder 53 has, for example, a configuration similar to that shown in FIG. The gain of the multiplier 22 is controlled by the value obtained by integrating the obtained data through the Silavik filter 25, and
The original analog audio signal waveform is reproduced by creating a staircase waveform as shown in FIG. 2, passing it through an analog switch 55 (described later), and removing unnecessary high-frequency components through a low-pass filter 26.
The clock generated by the clock oscillator 54 is provided as a reading clock to the memory 51 via the controller 52, and is further provided to the shift register 53.
The signal is supplied to the shift register 23 as a demodulation clock. The controller 52 controls the input of digital data to the decoder 53, the operation of the analog switch 55, etc. in the following manner. FIG. 6 shows an example of the storage format in the memory 51. As shown in FIG. )
Each address is stored in association with address 62.
Here, L indicates a data area of a silent section, and M indicates an area of data of a sound section. On the other hand, in the memory 51, as shown in FIG. 6b, data 63 indicating addresses K+3, K+i, . In order to create the table shown in FIG. 6b, when the encoder encodes the audio signal waveform, each time the start point of the voiced section of the audio signal waveform is detected, the address where the encoded data is written is set. M
You can write them in order starting from the address. In that case, it is necessary to discriminate between a sound section and a silent section of the audio signal waveform, and a circuit as shown in FIG. 7 may be used for this purpose. That is, the level of the audio signal 71 is detected by a level detector 72 whose rising and falling time constants are optimally set, and the detection output is passed through a comparator 73 whose threshold value V _th is optimally set. A presence/silence discrimination signal 74 as shown in FIG. 3b can be obtained. The controller 52 normally sends the encoded data 61 shown in FIG. 6a in the memory 51 to the decoder 53 as is, but the encoded data 61 shown in FIG.
From the data 63 indicating the starting point address of the sound section shown in
The start address of the sound section of the encoded data 61 to be read next, for example K+3, is known, and from that address τ
Calculate the address where the data of the silent section before the time is written. For example, if τ = 4 ms and the sampling frequency is 16 kHz, the address where the data of the silent period τ hours before address K+3 is written is 16 x 4 = 64 bits before address K+3, that is, 8 bytes. The previous address is K-5. Then, at the timing to read the contents of this address K-5 from the memory 51, the controller 52 reads out the encoded data (silent section data) stored in the memory 51.
The transmission to the decoder 53 is stopped, and instead, specific digital data for making the quantization width in the decoder 53 larger than the minimum quantization width is sent to the memory 51 for a time τ, that is, the data at the sound interval start point address K+3. The data is supplied to the decoder 53 for a period from 1 to 2 until immediately before the read timing (until the timing at which data at address K+2 is to be read). Here, the specific digital data for making the above quantization width larger than the minimum quantization width is:
In this embodiment, this is data for activating the Miravic filter 25 in the decoder 53, in other words, it is data of three or more consecutive "1" or "0" bits. In general, if the quantization width is assumed to increase when consecutive M bits of digital input to the decoder become the same value, then if this digital data has the same value consecutively for M bits or more, then the quantization width increases. good. In this example, this digital data is
As shown in the following table, 8 bits of "1" and "0" are alternately consecutive.

[Table] In this table, the corresponding address represents the address in the memory 51 to be supplied to the decoder 53, which has been replaced with the replacement digital data shown on the right. In this way, the output waveform of the multiplier 22 when the silence pattern during the period of time τ in the silent interval is replaced with digital data consisting of 8 consecutive bits of "1" and "0" and supplied to the decoder 53 is calculated as follows. It is shown in Figure 8a. As can be seen from this figure, within the period of τ, the quantization width △ is larger than the quantization (minimum quantization width) △ ₀ when the silence pattern is input to the decoder 53, and the sound interval start point The quantization width at ₀ is set sufficiently large. Therefore, the reproduced audio signal waveform obtained by passing the output of the multiplier 22 through the low-pass filter 26 closely follows the changes in the original audio signal waveform even in the early part of the sound section. That is, by compensating for the slope overload at the beginning of a voiced section, which has been a problem with adaptive delta modulation, during playback, the phonology becomes clearer and the intelligibility is greatly improved. By the way, if the waveform of FIG. 8a is directly passed through the low-pass filter 26, the waveform of the period τ will appear as noise. Therefore, since the period τ is a silent section, the analog switch 55 provided on the input side of the low-pass filter 26 is switched to the reference potential point E side of OV using the control signal shown in FIG. 8B. In this way, the input waveform of the low-pass filter 26 becomes 0 level in terms of direct current during the period τ as shown in FIG. 8c, and the output of the low-pass filter 26 becomes a silent level, so that no noise is generated. Note that the quantization width at the start point t ₀ of the sound section changes depending on the time τ or the content of the digital data supplied to the decoder 53 during the period τ.
However, if the quantization width is increased, the slope overload will be reduced, but the quantization noise will also increase, so it is necessary to take this point into consideration. However, in cases such as when a voiced interval begins with a consonant, the larger the quantization noise, the better the driving sound source effect of the consonant, and the phonology becomes clearer. This invention can be implemented with various modifications. For example, as the digital data for increasing the quantization width, appropriate coded data of the sound section in the memory 51 may be used. Further, the time τ for replacing this digital data with the encoded data of the silent section in the memory 51 may be set to the entire period of the silent section. Further, the analog switch 55 may be replaced with a voltage-controlled variable gain amplifier whose gain is controlled to be zero or sufficiently small during the period τ. On the other hand, as for the storage format in the memory 51, the data in the silent section is "1" and "0".
Instead of memorizing alternating patterns of silence,
A code indicating the generation timing of the silent section and data indicating the length of time may be stored. In this way, the capacity of the memory 51 can be saved. Further, in the embodiment, an example in which the present invention is applied to the CVSD method has been described, but the present invention reads out digital data including data encoded by adaptive delta modulation other than the CVSD method from the memory and converts it to the original voice. It can also be applied to devices that reproduce signal waveforms.

[Brief explanation of the drawing]

Figures 1a and b are diagrams for explaining slowo overload and granular distortion in the delta modulation method, and Figures 2a and b are CVSD
Figures 3a and 3b are diagrams showing an example of the envelope waveform of a voice signal and the corresponding waveform of the speech-no-speech discrimination signal. Figure 4 is a diagram showing an example of the envelope waveform of a plosive. FIG. 5 is a diagram showing some examples of waveforms, FIG. 5 is a diagram showing the configuration of an audio reproduction device according to an embodiment of the present invention, and FIG.
7 is a diagram showing an example of a storage format in a memory, FIG. 7 is a diagram showing an example of the configuration of a sound/non-sound discrimination circuit, and FIG. 8 is a time chart for explaining the operation of the same embodiment. 51...Memory, 52...Controller, 53
...Decoder, 54...Clock oscillator, 55...
...Analog switch.

Claims (1)

[Claims] 1. A memory that stores digital data including encoded data obtained by encoding an audio signal waveform using adaptive delta modulation, and a decoder that reproduces the original audio signal waveform from the digital data read from this memory. a means for detecting a silent section of digital data read from the memory; and a means for detecting a silent section of digital data read from the memory; An audio reproducing device comprising: means for supplying specific digital data; and means for maintaining the output of the decoder at a silent level while the specific digital data is being supplied to the decoder. 2. The decoder increases the quantization width when consecutive M bits of input digital data become the same value, and the specific digital data supplied to this decoder during part or all of the silent period is 2. The audio reproduction device according to claim 1, wherein the same value continues for M bits or more.