JPH02275996A - Word recognition system - Google Patents

Abstract

PURPOSE:To secure a high recognition rate and to easily perform real-time processing by using time variation of frequency characteristics of a voice as an input to a neural network. CONSTITUTION:The time variations 10 and 15 of the frequency characteristics of the voice are used as the input to the neural network 20. Thus, 'time variation of frequency characteristic of voice' is used, so preprocessing for obtaining the input is simplified and the time required for the preprocessing may be short. Further, the arithmetic processing of the whole neural network 20 is simple and speedy in principle. Consequently, the high recognition rate is secured and the real-time processing is easily performed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a word recognition method suitable for recognizing words from input speech using online terminals such as electric locks and IC cards.

[Prior Art] A conventional word recognition method, as described in, for example, Japanese Patent Publication No. 63-4200 and Japanese Patent Application Laid-open No. 62-220998, is based on the following procedure.

■Extract features related to words included in input speech.

(2) Calculate the distance between the standard pattern extracted in advance in the same manner as (2) above and the feature amount extracted in (2) above.

(2) Based on the calculation results, determine the word of the standard pattern with the minimum distance above as the word of the input speech.

[Problems to be Solved by the Invention] However, the conventional word recognition method described above has the following problems.

■The recognition rate deteriorates as time passes from the time the standard pattern was created.

■Real-time processing is difficult. In other words, in order to ensure a recognition rate above a certain level in conventional word recognition methods, it is necessary to use complex features, but extracting complex features requires a complex processing device, which reduces processing time. It will also be huge.

An object of the present invention is to obtain a word recognition method that ensures a high recognition rate and that can be easily processed in real time.

[Means for Solving the Problems] The present invention as set forth in claim 1 is a word recognition method that uses a neural network to recognize a word from an input voice. This method uses the temporal change in .

According to a second aspect of the present invention, a temporal change in an average frequency characteristic of audio within a certain period of time is used as an input to the neural network.

According to a third aspect of the present invention, the neural network is a hierarchical neural network.

[Function] According to the present invention as set forth in claim 1, there are the following functions and effects.

■Deterioration of recognition rate over time is extremely small. This means that
It is presumed that this is because it is possible for the neural network to adopt a structure that is less susceptible to fluctuations due to differences in speech timing.

■Since "temporal changes in the frequency characteristics of audio" are used as input to the neural network, the preprocessing to obtain the input is simpler than the conventional complex feature extraction. The time required is short and sufficient.

■Neural networks are, in principle, simple and quick to perform calculations on the entire network.

■In principle, each unit that makes up a neural network operates independently, and parallel arithmetic processing is possible. Therefore, calculation processing is quick.

(2) With the above (2) to (2), word recognition processing can be easily performed in real time without using a complicated processing device.

Further, according to the present invention as set forth in claim 2, in addition to the effects (1) to (2) above, there is the following effect (2).

■Since "temporal changes in the average frequency characteristics of audio within a certain period of time" are used as input to the neural network, the processing in the neural network is simple and the time required for this processing is shorter.

Further, according to the present invention as set forth in claim 3, in addition to the effects (1) to (2) above, there is the following effect (2).

■For hierarchical neural networks, currently,
A simple learning algorithm (pack propagation) as described below has been established, and a neural network that can achieve a high recognition rate can be easily formed.

[Example] Fig. 1 is a schematic diagram showing an example of a word recognition system to which the present invention is applied; Fig. 2 is a schematic diagram showing input speech; Fig. 3 is a schematic diagram showing the output of a bandpass filter; Figure 4 is a schematic diagram showing a neural network, Figure 5 is a schematic diagram showing a hierarchical neural network, and Figure 6 is a diagram showing a unit I.
FIG.

Prior to describing specific embodiments of the present invention, the configuration of the neural network and the learning algorithm will be described.

(1) Due to its structure, neural networks are
It can be roughly divided into two types: a hierarchical network shown in FIG. 4(A) and an interconnected network shown in FIG. 4(B). Although the present invention may be configured using either of these networks, the hierarchical network is more useful because a simple learning algorithm as described below has been established.

(2) Network Structure A hierarchical network has a hierarchical structure consisting of an input layer, an intermediate layer, and an output layer, as shown in FIG.

Each layer is composed of one or more units. The connections are forward-looking, from the human power layer to the intermediate layer to the output layer, and there are no connections within each layer.

(3) Structure of the unit The unit is a model of a neuron in the brain and has a simple structure as shown in FIG. It receives input from other units, sums it up, transforms it using a certain rule (conversion function), and outputs the result. Each connection with another unit is given a variable weight that represents the strength of the connection.

(4) Learning (pack propagation) Learning of a network is to bring the actual output closer to the target value (desired output). Learn by making changes.

Further, as a learning algorithm, for example, Rumel
hart, D.E., McClelland, J.
L, and thePDP Re5earch Gr.
oup, PARALLEL DISTRIBUTED
PROCESSING, the MIT Press
, 1986. The backpropagation described in (Reference 2) can be used.

Hereinafter, specific examples of the present invention will be described. still,
The recognition system 1 of this embodiment includes an n-channel bandpass filter 10. Consisting of a combination of an averaging circuit 15, a neural network 20, and a determination circuit 30 (first
(See figure) (A) The learning words are ``Shomei'', ``Air conditioner'', ``Curtain'', ``TV'', and ``Door'', and the input words are ``Shomei'', ``Air conditioner'', ``Curtain'', and ``TV''. , the five words were ``door''.

(B) Preprocessing ■ Input speech (5 words) is temporally equally divided into four blocks as shown in FIG.

■As shown in Figure 1, the audio waveform of each block is multiple (n
(in this embodiment, n=8) channels, and frequency characteristics as shown in FIGS. 3A to 3D are obtained for each block, that is, for each fixed time period.

At this time, the output of the bandpass filter 10 is averaged by an averaging circuit 15 for each block.

(C) Processing and judgment by neural network ■Results of preprocessing (Bant pass filter 10 for each block
output) is input into a three-layer hierarchical neural network 20 as shown in FIG. The input layer 21 is composed of 4×n units (n=832 units in this embodiment) corresponding to 4 blocks of preprocessing and n channels. The output layer 22 has two units corresponding to registered words and other words for each of the five words,
The whole is composed of 10 units.

- The output of the neural network 20 is input to the judgment circuit 30, and the words of the input speech are recognized. However, in implementing the present invention, the output of the neural network 20 is not mechanically judged by the judgment circuit 30, but is judged by the human intellect that has obtained the output of the neural network 20. Also good.

■By backpropagation of the learning algorithm described above, the network is trained 1,000 times until the error in the output relative to the input converges to a certain level, and a network that can guarantee a certain recognition rate is constructed. It should be noted that learning was performed so that the output would be "1" for the unit corresponding to each registered word, and "0" for the units corresponding to other words.

(D) Experiment Word recognition was experimented using the recognition system 1 described above.

The input speech was learned using pack propagation.
Words ("shomei", "air conditioner", "curtain",
``television'', ``door'').

(a) Recognition rate As a result, the recognition rate was found to be 100%.

(b) Processing speed Also, the processing speed (the time required to recognize the utterance of one word) was within 1 second, and it was recognized that extremely rapid processing could be performed.

That is, in the above-mentioned recognition system 1, the recognition rate is extremely high as shown in the result of (a).

Further, in the recognition system 1, as shown in the result (b) above, speaker recognition processing can be performed quickly without using a complicated processing device, and can be easily performed in real time.

In the recognition system 1 described above, the averaging circuit 15 is used to use the temporal change in the "average frequency characteristic over a certain period of time" of the voice as input to the neural network 20. In carrying out the invention,
It is also possible to simply use "temporal changes in frequency characteristics of audio" as input to the neural network.

[Effects of the Invention] As described above, according to the present invention, the word recognition method % formula % ensures a high recognition rate and easily performs real-time processing.

[Brief explanation of drawings]

Fig. 1 is a schematic diagram showing an example of a word recognition system to which the present invention is applied, Fig. 2 is a schematic diagram showing input speech, Fig. 3 is a schematic diagram showing the output of a bandpass filter, and Fig. 4 is a schematic diagram showing an example of a word recognition system to which the present invention is applied. FIG. 5 is a schematic diagram showing a hierarchical neural network, and FIG. 6 is a schematic diagram showing the structure of a unit. DESCRIPTION OF SYMBOLS 1... Recognition system, 10... Band pass filter, 15... Averaging circuit, 20... Neural network, 21... Input layer 22... Output layer, 30... Judgment circuit ( word recognition section). Patent applicant Sekisui Chemical Co., Ltd. Representative Hiroshi 1) Kaoru Toda Δ゛ (Yo, Ki-

Claims (3)

[Claims] (1) A word recognition method that uses a neural network to recognize a word from input speech, and uses temporal changes in the frequency characteristics of the speech as input to the neural network. (2) The word recognition method according to claim 1, wherein temporal changes in average frequency characteristics of speech within a certain period of time are used as input to the neural network. (3) The word recognition method according to claim 1 or 2, wherein the neural network is a hierarchical neural network.