JPH04323700A - Voice signal recording device - Google Patents

Abstract

PURPOSE:To obtain the voice signal recording device which performs efficient encoding. CONSTITUTION:A filter divides PCM data to be recorded into signals in four frequency bands. A power calculation part 12 calculates power, frame by frame, by the divided frequency bands and an assigned bit calculation part 50 calculates the numbers of assigned bits, frame by frame, in the respective bands from the power. Then the voice signals in the respective bands after being normalized corresponding to the power are quantized with the numbers of assigned bits and outputted as data c1-c4. Then p1-p4 by the frames and c1-c4 by words are stored in a memory. Therefore, the p1-p4 are the data by the frames and data by the words corresponding to the assigned bits determined by c1-c4 and p1-p4. Therefore, the optimum encoding is performed, frequency by frequency, without increasing the number of data.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio signal recording device for efficiently compressing audio signals and recording them in a semiconductor memory or the like.

0002

2. Description of the Related Art Conventionally, various audio recording devices have been used, and magnetic recording is the most widely used. on the other hand,
With recent advances in semiconductor technology, semiconductor memories have become cheaper and are now being used in audio signal storage devices. Audio signal recording devices using this semiconductor memory have the advantage of being able to perform high-speed and random access, and do not require a mechanical drive mechanism for recording and playback, so they are widely used in answering machines, etc. has been done.

However, semiconductor memory is more expensive than magnetic memory, and when audio signals are recorded therein, it is desirable to compress the signals as much as possible. Since the signals stored in the semiconductor memory are digital signals,
Various encoding and quantization techniques are used to reduce the amount of data compared to simply converting an analog audio signal to a digital signal. That is, there have conventionally been methods that combine PCM, differential PCM, and quantization to compress audio signals. In addition, in such PCM encoding, AB (Adaptive Bit A) changes the allocation of quantization bits according to the data content.
location) etc. are also known.

0004

[Problems to be Solved by the Invention] However, conventional compression techniques perform code compression in the direction of the amplitude of the audio signal, and therefore, in order to further improve memory usage efficiency, the sampling frequency must be set low. I will do it. And when the sampling frequency is lowered,
The recorded data becomes discontinuous, and the reproduced audio data becomes distorted. This distortion occurs particularly in high frequency regions. Therefore, as an encoding means to solve this problem, SBC (Sub
Band Coding) is known. This SB
In C, the codes of each band are transferred to a common baseband by a decimation technique to reduce the required sampling rate. However, this SBC has the problem that the total number of bits required increases because each band is encoded separately.
It was not possible to record efficiently.

The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to provide an audio signal recording device that can improve memory usage efficiency.

[0006]

[Means for Solving the Problems] An audio signal recording device according to the present invention includes a frequency filter that divides an input digital audio signal into signals of each of a plurality of frequency bands, and a power of the signal of each divided frequency band. and quantization means that allocates the number of bits according to the power of the signal in each frequency band and quantizes the signal in each band.

[0007]

[Operation] A digital audio signal is separated into signals for each of a plurality of frequency bands by a frequency filter. Then, the power is calculated for each separated frequency band, and the number of bits to be allocated during quantization is determined according to the result. Therefore, the number of bits allocated to each band can be optimized, and efficient encoding processing can be performed.

[0008]

Embodiments Hereinafter, embodiments of the present invention will be described based on the drawings.

Recording unit configuration Recorded PCM obtained by PCM encoding an analog signal input from a microphone or output from another device.
The audio signal is divided into four frequency band signals a1, a2, a3, and a4 by two-stage (three) QMF filters (band division filters) 10a, 10b, and 10c, and is low-frequency converted to baseband. be done. This QMF filter sampling 10 separates the recorded PCM signal into a high frequency band and a low frequency band at a frequency of 1/4 of the sampling frequency, and downsamples it to 1/2 of the sampling frequency. The frequency band is divided into four frequency bands.

These divided signals a1 to a4 are input to power calculation units 12-1 to 12-4 and n-word memories 14-1 to 14-4, respectively. The n-word memory 14 stores n pieces of data a1 to a4 for each word. Here, n is the number of data in one frame, and normally one frame is a signal amount of about 10 msec. The reason why one frame is about 10 msec is because the human voice repeats the strength and weakness at a certain period (pitch period), and this period is about 10 msec.
This is because the time period is often about 0 msec, and it is preferable to divide the signal in accordance with this time period. On the other hand, the power calculation section 12
calculates the powers p1, p2, p3, p4 for each band from the values of the signals a1 to a4 for each band for each frame, and quantizes the calculated powers p1 to p4 in the form of 22σi. Output. Here, σi is an integer, and i is a variable representing the band (1 to 4 in this example). In addition, in reality, the obtained power is expressed in binary numbers, and the leftmost (
σi (σ1 to σ4) is obtained by searching for a 1 (higher side), finding out how many bits from the right it is, and subtracting 1 from this value and dividing by 2.

[0011]The powers p1 to p4 obtained in the power calculation unit 12 are then applied to the corresponding normalization units 16-
1 to 16-4. This normalization unit 16-1 to 1
6-4 is data a output from the n-word memory 14.
1 to a4 are normalized so that the data power in each band is 1. Actually, the signals a1 to a4 of each band
is the square root of the quantized power in that band (i.e. 2σ)
Normalize by dividing by. Here, in order to accumulate data for one frame in the n-word memory 14, based on the power calculated for that frame,
Can be normalized. The normalized data b1 to b4 are then supplied to quantization units 22-1 to 22-4.

On the other hand, the powers p1 to p4 for each band obtained by the power calculation units 12-1 to 12-4 are supplied to the allocated bit calculation unit 20. Then, the allocated bit calculation unit 20 calculates the allocated bit numbers m1, m2, m3, and m4 for each band in the following manner.

Ri=Rav−Wi+σi −Σwj σj (j
=1~N) Wi=(1/2)log2 (wiΠw
j - wj ) (j = 1 to N) where Rav is the average bit rate per sample, Ri is the number of bits allocated per sample in each band, wi is the bandwidth ratio of the i band, and Wi is a constant.

In this example, since wi=1/4, the above equation becomes Ri=Rav+σi −(σ1+σ2+σ3+σ4)/
4 (in addition, wiΠwj−wj = 1/4 {(1/4)−
1/4*(1/4)-1/4*(1/4)-1/4*(
1/4)-1/4}=(1/4)*4=1, so
Wi is 0. ) becomes. Therefore, the allocated bit calculation unit 20 calculates the number of bits m for each of the four bands based on the calculation result obtained in this way.
1 to m4 are determined and supplied to the quantization units 22-1 to 22-4. Therefore, each quantization section 22-1 to 22-4 compresses the input data according to the assigned number of bits and outputs it as compressed data c1 to c4 for memory storage. Further, the powers p1 to p4 for each band are also necessary to determine the number of allocated bits during decoding, and these are also output.

Data format in memory Next,
The data format of the signal recorded and reproduced in the memory will be explained based on FIG. 2. As mentioned above, the recording unit outputs p1, c1, p2, c2, p3, c3 for each band.
, p4, c4 are output. One frame of data consists of n words. On the other hand, powers p1 to p4
is determined for each frame, and for data of one frame, the allocated bits for each band are the same. Therefore, for m bits corresponding to entry 0 of one frame of data, p1 is set as data of m/4 bits each.
~p4 data is recorded. Then, data c1 to c4 are stored in the next entries 1 to n, respectively. The data c1 to c4 are calculated for each word as described above, and n pieces of data are recorded. Therefore, one frame consists of n+1 words. And for the next frame,
p1 to p4 obtained for the frame are recorded, and n words of c1 to c4 are recorded corresponding to the allocated bits determined based on p1 to p4.

Configuration of Reproducing Side Next, reproduction of data recorded as described above will be explained based on FIG. 3. c read from memory
Data 1 to c4 are supplied to decoding units 40-1 to 40-4. On the other hand, data p1 to p4 regarding the power calculated and recorded for each frame are supplied to the allocated bit calculating section 50. This allocated bit calculation unit 50 is
The numbers m1 to m4 of allocated bits are calculated by the same calculation as the allocated bit calculating section 20 described above, and are supplied to the decoding sections 40-1 to 40-4, respectively. Therefore, these decoding units 40-1 to 40-4 dequantize the input data c1 to c4 according to the number of allocated bits, and restore the data to pre-quantization data b1 to b4. And this data b1~b
4 is supplied to the power recovery units 60-1 to 60-4. The power return units 60-1 to 60-4 are supplied with the read powers p1 to p4, respectively, and the data a1 to a4 are multiplied by the powers p1 to p4 to the data b1 to b4, respectively. Play each. These data a1 to a4 are integrated in two stages by QMF filters 70a, 70b, and 70c, respectively, to form a reproduced PCM signal.

As described above, according to this embodiment, the input audio signal is divided into signals of four frequency bands. Then, the power is calculated for each divided frequency band signal, and the allocation of the number of quantization bits is determined based on this. The number of quantization bits is switched every frame. Therefore, the power calculation for determining the number of quantization bits can be made accurate, and the optimal number of quantization bits can be allocated. Also, 1
Since the power need only be stored for each frame, the area for power storage can be saved.

Overall Configuration of Audio Signal Recording and Reproducing Apparatus Next, the overall configuration of the recording and reproducing apparatus will be explained based on FIG. In the figure, an analog-to-digital converter ADC converts an analog recorded signal into a recorded PCM signal. This data is then supplied to a high-speed signal processing device DSP. Also,
The reproduced PCM signal from this DSP is converted into an analog reproduced signal by a digital-to-analog converter DAC.

[0019] The high-speed signal processing device DSP is a
Data p1, p2, p3, p4 and c1, c2, c3, c4 for memory recording are obtained from the function shown in FIG.
The above-mentioned encoding process is performed on the recorded PCM signal for each frame. The encoded data p1 to p4 and c1 to c4 are then supplied to the audio memory MEM via the multiplexer MPX and the memory interface MEMIF. Then, in this audio memory MEM, recording is performed in the format based on the above-described FIG. 2. On the other hand, the playback data read from the audio memory MEM is transferred to the memory interface ME.
The signal is supplied to the high-speed signal processing device DSP via the MIF and multiplexer MPX, and the high-speed signal processing device DSP outputs a reproduced PCM signal using the function shown in FIG. 3 described above. Then, this reproduced PCM signal is converted into an analog reproduced signal by a DAC.

Here, these operations are controlled by an operating mode controller CON. That is, when writing to the audio memory MEM, the mode controller CON supplies a control signal (memory write MW) and a memory address to be recorded via the memory interface MEMIF. Therefore, memory interface ME
Recorded data supplied from the MIF to the audio memory MEM will be recorded at a predetermined address according to the above format. On the other hand, when reading playback data,
The operating mode controller CON supplies a control signal (memory read MR) and a signal regarding the address to be read via the memory interface MEMIF. Therefore, the data at the corresponding address in the audio memory MEM is supplied to the high speed signal processing device DSP via the memory interface MEMIF and the multiplexer MPX on the data bus as playback data. The above-described reproduction operation is performed by this high-speed signal processing device DSP. Here, the operations of the memory interface MEMIF, multiplexer MPX, and high-speed signal processing device DSP are also controlled by the mode controller CON. In this way, the audio signal is recorded and reproduced.

[0021]

As described above, according to the audio signal recording apparatus according to the present invention, recorded audio signals are divided into frequency bands and encoded. Then, the number of allocated bits for this encoding is determined based on the power value for each band. Therefore, efficient encoding can be performed as a whole, and long-time recording and playback can be performed using a small memory capacity.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of an audio signal recording device according to the present invention.

FIG. 2 is a format diagram of recording data in the same embodiment.

FIG. 3 is a block diagram of an optimally suitable reproduction device for data recorded according to the present embodiment.

FIG. 4 is a block diagram of an audio signal recording and reproducing apparatus to which this embodiment is suitably applied.

[Explanation of symbols]

12 Power calculation section 14 n word memory 16 Normalization section 20 Allocated bit calculation unit 22 Quantization section

Claims (1)

[Claims] Claims 1: A frequency filter that divides an input digital audio signal into signals for each of a plurality of frequency bands; power calculation means that calculates the power of the signal of each divided frequency band; An audio signal recording device comprising: quantization means for allocating the number of bits according to the power and quantizing the signal of each band.