JPS6269298A - Voice recognition equipment - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION (A) Field of Industrial Application The present invention relates to a speech recognition device in which the state of an identification target in the next recognition process changes depending on the result of speech recognition.

(B) Prior Art Generally speaking, the recognition rate of speech recognition apparatuses decreases as the number of recognized words increases, as shown in Japanese Patent Application Laid-Open No. 59-219798. Conventionally, in order to prevent this, the recognition rate was improved by dividing the recognition words into several groups and limiting the words to be recognized the next time according to the result of speech recognition. For example, when controlling electrical appliances using voice recognition, first the voice ``light'' is recognized, then only the three words ``on'', ``off'', or ``cancel'' are recognized. I can do it. Figure 2 shows an example of such state transition rule information, and shows the state transition rule information for each state (state 1, state 2, state 6, etc.) and the state transition rule information for recognition &ik vs. unspoken and recognition results. Transition information is stored in gd. For example, let's say you want to control a light, a cooler, and a timer attached to the cooler using voice recognition. The 15 words used are the numbers ``ichi'' to ``kyu'', ``light'', ``cooler'', ``timer'', ``on'', ``off'', and ``trikeshi''. The symbol X in the figure means that the words are not recognized, and the numbers indicate the number of the seven bales in the first order that changes when each word is recognized.

Assume that the cooler is currently stopped, and this is set to state 1.
shall be. In Case 41, the only words to be recognized are ``light'' and ``cooler.'' First, if the voice "light" is input and recognized, a transition is made to state 2. In state 2, any one of "on", "off", and "trigger" is recognized, and the state returns to state 1. When "on" or "off" is recognized, a predetermined signal is output from the output section. Here, the state is the same when the light is on and when it is off! However, if it is inconvenient for a signal indicating that the light is on to be output when the light is on, another state may be defined.

On the other hand, if it is recognized as a "cooler" in state 1, the state transitions to state 6. Here, only "ON" and "TRIKESH" are to be recognized, and if "ON" is recognized, the process moves further to step 7 and the cooler starts. Once the cooler starts, the timer can be set, and the three words to be recognized are "light,""cooler," and "timer."

If "timer" is recognized in state H4, the state moves to state 7 and the words to be recognized are numbers from "ichi" to "kyu", "off" and "trick". When the number is recognized, a signal is output to set the time until the cooler stops, and when "off" is recognized, a signal is output to cancel the timer setting.
1ii171. In this case, the state returns to state 4 as well.

Conventionally, in order to memorize and control such state transition rules, the speech recognition device is controlled by a host computer, and the next recognition target word is notified to the speech recognition device based on the recognition result output from the speech recognition device. has been used. However, since this method requires a host computer, the system tends to be complicated and large.

On the other hand, the system can be simplified by storing the state transition rules in the speech recognition device and having the speech recognition device itself monitor the state transitions and determine the recognition target words. Figure 5 shows a diagram of such a system. It is a block diagram showing a configuration example, where il+ is a microphone, (2) is a feature extraction unit that extracts features of input audio, and (3) is the feature extraction unit (
This is a pattern creation section that creates an input pattern using the special @ of the input voice obtained in step 21. (4) is a standard pattern storage section that stores a pre-registered number of standard patterns; (5) is a standard pattern storage section (
4) is a pattern matching section that compares the pattern with the standard pattern (61 is an output section that outputs a predetermined signal to the outside according to the result of pattern matching. (71 is a state as shown in FIG. 2) , the state J4 transfer rule storage unit for storing the S rule, (8) is the state jfA transfer rule storage unit!71 f
A standard pattern specifying section (91) selects a standard pattern in the standard pattern storage section (4) using J'Mi; ) is a gate that selects the standard pattern to be sent to.

However, in the past, the state transition rules were incorporated as part of the program of the microprocessor that controlled the speech recognition device, which meant that the speech recognition device was specialized, and the program for the speech recognition device had to be rewritten depending on the purpose. (c) Problems to be solved by the invention The present invention is intended to solve all the disadvantages of upper IQ.
It is an object of the present invention to provide a speech recognition device that can change the state transition rules stored therein by simple means.

(b) Means for solving the problem A feature extraction unit that extracts special@tsuchi from the input voice, identifies the input voice based on the extracted features, and generates a predetermined signal according to the result. an identification unit that outputs a dpi identification unit; and a desired transition rule storage unit that stores a transition rule between states of the number of stages related to an identification target that transitions according to the identification result of the previous dpi identification unit;
The state transition rule storage section is characterized in that the state transition rule storage section has a completely blind structure that is removable from the identification section.

Furthermore, the present invention is characterized in that the state transition #rule storage unit is constituted by a ROM (read only memory) installed in the IC socket operator. Since the identification unit is configured as follows, the state transition rule 1] can be easily changed by simply exchanging the state transition rule μ(j memory unit, for example, 'FICM'i) by determining all the objects to be identified according to the rules of the It is possible to add state transition rules to the speech recognition device according to the purpose.

(v) Embodiment FIG. 1 shows an example of the present invention, and in order to embody FIG.
, was realized by replacing the stacking pattern specifying section (8) and gate (9) with a microprocessor and arithmetic circuit, and those having the same function as the second haze are given the same number as 1-. . The audio input through microphone +11 is amplified by amplifier 8 (101,1), and then passed through BPF (Bant filter).
pass filter) (divided into et number bands m by IIJ,
It is digitized by an ADO (Analog-Digital Converter) (121) to create a microgrosse pattern. When registering a voice, a specific number of patterns are stored in the standard pattern storage unit (4) as the standard pattern for the large input cover, and the voice During recognition, the input cover pattern and the standard pattern storage section (
The microlog setter (131) compares each standard pattern stored in the microlog setter (4) using the arithmetic circuit α41, and the result is outputted to the outside through the output unit (6).

On the other hand, the microlog 0 setter (13 is the state B transition rule stored in the state transition rule storage section (7)) is stored in the previous sound PJ.
The target word to be recognized is judged while referring to the fi result.

The difference between the device according to the embodiment of the present invention and the conventional device is in the state transition rule storage section (71).
consists of a ROM (711) in which state transition rules are stored and an IC socket σ2 for mounting the ROM, and the ROM can be easily replaced.

In this way, the 'ROM key 1 of the state transition rule storage section (7)
] is hardware-separated from the program storage section σ9, and corresponds to the purpose of the device, and can be replaced with another ROM that stores the 4-secret rule table in this state, so there is no need to modify the program. The uses of the device can be greatly increased.

(G1) Effects of the Invention As is clear from the above explanation, according to the present invention, the simple Mizukuri/state transition rule can be used without the need for a host computer, and the state transition rule can be easily created. Since it can be changed, it is possible to provide an excellent speech recognition device that has a high recognition rate and can be used for many applications.

[Brief explanation of drawings]

The first mark is the voice recognition of the present invention, ? FIG. 2 is a schematic diagram of a four-body example of state transition rules, and FIG. 5 is a block diagram of a voice recognition device using state transition rules. +11-...Microphone, (4)--Standard pattern storage unit, (7i'...State transition rule storage unit, (131
-Microprocessor, ■...Arithmetic section, Vt.
・ROM. Blink...Socket

Claims (2)

[Claims] (1) A feature extraction unit that extracts features from input audio, and identifies the input audio from among many recognition targets based on the extracted features, and outputs a predetermined signal according to the result. and a state transition rule storage unit that stores transition rules between a plurality of states regarding an identification target that changes according to the identification result of the identification unit, and the identification unit is detachable from this. A speech recognition device characterized in that a recognition target is determined according to a rule in the state transition rule storage unit that is connected in a series of states, and then a recognition process is performed. (2) The voice recognition device according to claim 1, wherein the state transition rule storage unit is comprised of a ROM (read-only memory) that is removably installed on an IC socket.