JP2507311B2 - Voice analyzer - Google Patents

Description

Description: 1) Field of the Invention The present invention relates to voice recognition and voice analysis by digital processing.

2) Prior Art Generally, when performing voice recognition and voice analysis by digital processing, a voice signal having a huge amount of information is orthogonally transformed by Fourier transform, Walsh transform, or the like to compress the amount of information,
Extraction of feature quantities is performed.

In the case of using the Fourier transform or Walsh transform, it is general that a time window is provided in consideration of the frequency resolution and the time variation of the feature amount, and a part of the voice that is time series data is extracted and processed.

This time window is often about 32 msec due to the balance between the sampling interval of the voice signal and the ease of processing (references, for example, Kyoritsu Shuppan "Speech Recognition" by Yasunaga Niimi, Ohmsha "Speech Information" Basics of processing "Sato Saito et al., Etc.).

The time window length is, for example, a sampling interval of 0.125 mse.
In the case of c, since there are 256 data in terms of the number of data, 256 storage areas for holding this data are required.

Further, in order to improve the time variation followability of the feature amount of the audio signal, it is widely practiced to partially overlap the time windows for calculation. For example, it is common to overlap 16 msec of 32 msec. The appearance time interval of the processed data is called a frame rate.

In order to realize a processing device that satisfies the above conditions, at least 384 or more storage areas and 16 mse
Since it is required to have a high speed so that it can be completed within c, it has hitherto been impossible to realize with a device using a general-purpose one-chip microprocessor which is inexpensive, has a small storage area, and is relatively slow.

3) Object of the Invention The present invention is to provide a speech analysis apparatus which can be realized at a low cost by reducing the storage area and the calculation amount which are the above-mentioned drawbacks.

4) Configuration of the invention In a voice analysis device that performs real-time processing, a voice analysis unit that is shorter than a conventional time window, that is, has a smaller storage area and less processing amount, a buffer unit that holds an analysis result, and one or a plurality of units. Adder, a buffer unit that holds the addition result that forms a pair with the adder, a control unit that controls the adder and the buffer unit, an adder that adds the contents of the buffer unit, and the addition result Is characterized in that it has a buffer unit for holding

5) Effect of the Invention According to the present invention, a voice analysis device that has conventionally required a large storage capacity and a large amount of processing can be used to obtain an equivalent amount of information with a small storage capacity and a small amount of processing. The device can be realized.

For example, the above example (analysis time window 32msec, frame rate 1
6msec, sampling interval 0.125msec), the memory capacity is 3
Assuming 84 pieces, the amount of processing is 1024 butterfly calculations if FFT is used.

If this is performed by an FFT with a time window of 8 msec in the speech analysis unit, the storage capacity will be 64, and the processing amount will be 192 times of butterfly calculations, and the storage capacity will be 1/6 and the processing amount will be approximately 1/5. There is.

6) Example An example of the speech analysis apparatus of the present invention will be described below with reference to the drawings.

FIG. 1 is a main block diagram of a voice analysis device corresponding to an analysis time window of 32 msec and a frame rate of 16 msec when the time window of the voice analysis unit is 8 msec.

 The voice analysis unit may be of a known analysis method.

In FIG. 1, the voice analysis unit 1 analyzes voice data of 8 msec by a known method and writes the analysis result in the analysis buffer 2. The voice analysis unit 1 also sends a signal to the controller 6 each time one analysis is completed.

The controller 6 switches the changeover switch 3 in the order of contact 12 → contact 13 → contact 14 → contact 15 → contact 12 according to a signal from the voice analysis unit 1. When the contact 12 is connected, the contents of the analysis buffer 2 are sent to the buffer 7. When the contact 13 is connected, the contents of the analysis buffer 2 and the contents of the buffer 7 are added by the adder 4 and sent to the buffer 7.

Connecting contact 14 sends the contents of analysis buffer 2 to buffer 8. When the contact 15 is connected, the contents of the analysis buffer 2 and the contents of the buffer 8 are added by the adder 5 and sent to the buffer 8.

When the contact 13 and the contact 15 are connected, 16 msec worth of data exists in the buffer 7 and the buffer 8.

Only when the contacts 13 and 15 are connected, the controller 6 sends a signal to the adder 9, adds the contents of the buffers 7 and 8, and sends the result to the divider 10.

In this case, the divider performs the averaging process of the analysis results for four times, and thus performs the process of dividing by 4.

The division by 10 can be replaced with a shifter because the process of dividing by 4 can be realized by a 2-bit shift by a shifter.

The division result is sent to the averaging buffer 11, and the averaging buffer 11 outputs the contents of the buffer as the analysis result.

 The time transition of the series of processes is shown in FIG.

In FIG. 2, timing 1 is the timing at which the analysis result is output from the voice analysis unit 1 in FIG.

 T is 8 msec in the above example.

Timing 2 is the timing at which the contact 12 and contact 14 in FIG. 1 are connected, and timing 3 is the timing at which the contact 13 and contact 15 in FIG. 1 are connected and the adder 9 operates and the analysis result is output. Is.

In the above embodiment, with respect to the time window T of the voice analysis unit,
Although an analyzer having a time window of 4T and a frame rate of 2T is shown, the number of contact points of the changeover switch 3 and the number of pairs of buffer and adder in FIG.
It is obvious that a configuration such as 6T and frame rate 3T can be used.

As described above, according to the voice analysis device of the present invention, a voice analysis device having the same amount of information as a conventional large storage capacity and large processing amount voice analysis device is realized even with a small storage capacity and a small processing amount. It is possible.

[Brief description of drawings]

FIG. 1 is a main block diagram of an analyzer according to an embodiment of the present invention, and FIG. 2 is a timing diagram of FIG. 1 ... Voice analysis unit, 2 ... Analysis buffer, 3 ... Changeover switch, 4 ... Adder, 5 ... Adder, 6 ... Controller, 7
... buffer, 8 ... buffer, 9 ... adder, 10 ... divider or shifter, 11 ... averaging buffer.

Claims (1)

(57) [Claims] 1. A voice analysis device for storing and outputting an analysis signal analyzed by a voice analysis unit in a predetermined amount of buffer,
A voice characterized by using a buffer having a capacity smaller than the predetermined amount of buffer, adding an analysis signal input to the buffer having a small capacity and a newly input analysis signal, and then performing averaging processing. Analysis equipment.