CN110910869B - Voice recognition method and device - Google Patents

Abstract

A speech recognition method includes the steps of grouping command words having complementary functions and similar parts in pronunciation into groups, setting first recognition threshold groups A1 and B1 and second recognition threshold groups A2 and B2, respectively; and a1 is less than a2, B1 is less than B2; and recognizing the command word, judging whether the command word threshold value is larger than any one of a first recognition threshold value group and a second recognition threshold value group A2 and B2 of any pair of command complementary words, if the command word threshold value is not larger than the second recognition threshold value group but larger than any one of the first recognition threshold value group, recognizing the current working state of the equipment, and switching the equipment to the complementary other state after recognition. The invention also discloses a voice recognition device suitable for adopting the method. The invention divides the complementary commands into groups and thresholds, and effectively controls the equipment by voice through recognition, thereby improving the recognition accuracy of the voice control command with high contact ratio.

Description

Voice recognition method and device

Technical Field

The invention belongs to the field of artificial intelligence, relates to a voice recognition technology, and particularly relates to a voice recognition method and voice recognition equipment.

Background

Many current devices are controlled by speech recognition, which uses a function command word or sentence that can express the function to speak to the device, and the device recognizes the speech function command word or sentence, understands the meaning and performs a corresponding control function. In practical application, a plurality of equipment functions such as startup, shutdown and the like have the control command words close to each other and only differ by one word, but have mutually exclusive functions, and when a controller speaks at a high speed or pronounces unclear, the equipment often causes misrecognition and executes a behavior inconsistent with the requirement of the controller, so that the experience of equipment use is greatly influenced.

Disclosure of Invention

In order to overcome the defects in the prior art, the invention discloses a voice recognition method and voice recognition equipment.

The voice recognition method comprises the following steps:

the command words with complementary functions and similar pronunciations are grouped and associated pairwise, wherein the complementary functions refer to that the equipment can only work in a non-A or B working state; and sets first recognition threshold groups a1 and B1, and second recognition threshold groups a2 and B2, respectively, for two command complements within the same group; and a1 is less than a2, B1 is less than B2;

recognizing the command word, judging whether the command word threshold value is larger than any one of the first recognition threshold value group A1 and B1 of any pair of command complementary words, if so, continuously judging whether the command word threshold value is larger than any one of the second recognition threshold value group A2 and B2;

if the current working state of the equipment is not greater than the second recognition threshold set but is greater than any one of the first recognition threshold set, the equipment is switched to the complementary other state after the current working state of the equipment is recognized;

if the recognition threshold value is not greater than any of the first recognition threshold value groups a1 and B1, the comparison recognition with the other command words is continued.

Preferably, the first recognition threshold sets a1 and B1 are set for similar parts in the command word.

The invention also discloses voice recognition equipment, which comprises a command word bank, a recognition module and an output module, wherein the command word bank comprises common words and associated words, every two of the associated words are associated, pronunciation of the associated words has similarity, the commands are complementary, the voice recognition equipment also comprises an equipment state detection module, and the recognition module is also in signal connection with the equipment state detection module.

Preferably, the identification module comprises a primary identification module and a secondary identification module, and the secondary identification module is in signal connection with the equipment state detection module.

The method divides the complementary commands into groups and thresholds, combines the control intention of a controller and the current state of the equipment, and performs effective voice control on the equipment through recognition, so that the recognition accuracy of the voice control command with high contact ratio is improved, and the hardware cost is not increased; has the advantages of obvious effect, convenience and easy use.

Drawings

FIG. 1 is a flow chart illustrating a speech recognition method according to an embodiment of the present invention;

fig. 2 is a schematic diagram of an embodiment of the speech recognition apparatus according to the present invention.

Detailed Description

The following provides a more detailed description of the present invention.

The basic principle of speech recognition is that after a sound signal of a command word is collected by a microphone and converted into an electrical signal, the electrical signal is compared with stored data, and the sound signal is recognized by comparison and an instruction corresponding to the sound signal is called to perform corresponding operation on equipment. For example, the command words issued by the controller's mouth are: and starting the air conditioner. After the air conditioner microphone with the voice recognition function reads the voice signal and converts the voice signal into an electric signal, the electric signal is compared with the stored data, and the air conditioner starts to work after the voice signal is recognized.

In the recognition process, no sound signal which is completely coincident exists, whether a command word is coincident with a certain command word stored in a command word bank or not is judged by setting a threshold, in the voice recognition, the recognition is carried out by judging the similarity of a pronunciation signal and a certain word in the existing word bank, the threshold can be the similarity, for example, an electric signal converted by reading the sound signal is scored according to the similarity of the signal and a certain word in the command word bank, the score is higher than a certain value, namely the word is considered to be coincident and a corresponding command is sent, and at the moment, the value is the threshold of the voice recognition.

The voice recognition method comprises the following steps of grouping and associating command words with complementary functions and similar pronunciations in pairs, wherein the complementary functions refer to that equipment can only work in a non-A or B working state; and sets first recognition threshold groups a1 and B1, and second recognition threshold groups a2 and B2, respectively, for two command complements within the same group; and a1 is less than a2, B1 is less than B2;

recognizing the command word, judging whether the command word threshold value is larger than any one of the first recognition threshold value group A1 and B1 of any pair of command complementary words, if so, continuously judging whether the command word threshold value is larger than any one of the second recognition threshold value group A2 and B2;

if the current working state of the equipment is not greater than the second recognition threshold set but is greater than any one of the first recognition threshold set, the equipment is switched to the complementary other state after the current working state of the equipment is recognized;

if not greater than any of the first recognition threshold sets A1 and B1, recognition continues in comparison to other command words.

The functions of the invention are complementary, command words with similar parts exist in pronunciation, and widely exist in practical operation, such as 'air conditioner on' and 'air conditioner off', 'air conditioner refrigeration' and 'air conditioner heating', 'microwave oven door opening' and 'microwave oven door closing', and the like, and the above command words can be seen from the meaning of the command words, the equipment can only work in a working state corresponding to non-A, namely B; the command words have the characteristic of high pronunciation similarity, and are easy to make mistakes in recognition.

One specific embodiment of the present invention for recognizing such words is as follows:

for example, for two words of air-conditioning refrigeration and air-conditioning heating, the words are firstly divided into the same group and are mutually associated, and two groups of thresholds A1 and B1, A2 and B2 are respectively set for the words, wherein A1 is smaller than A2, and B1 is smaller than B2; the a1 and the B1 can recognize the same similar parts of the two words in the group, the similar parts are the same pronunciations of the first three words for the two command words of air-conditioning cooling and air-conditioning heating, the two command words both comprise four words, and the like, and when the first three words, namely the null modulation X, are recognized and have the same sequence, and a fourth undetermined word X is also provided, the situation that the number of the words is larger than any one of the first group threshold value can be set.

After the controller sends out a command word, a primary recognition module in the recognition modules firstly recognizes whether the command word reaches a certain threshold value A1 or B1 of a first recognition threshold value set of a certain group of associated words, if the command word does not meet the certain threshold value, other command words which are not grouped or other grouped associated words are continuously recognized, if the command word meets the certain threshold value, a secondary recognition module continuously performs comparison recognition and probability scoring with a second recognition threshold value set A2 and a second recognition threshold value set B2 corresponding to two words in the group, and if the comparison result shows that the comparison result is larger than any final recognition threshold value, an instruction corresponding to the command word is executed. Fig. 1 shows a specific flow chart of the identification of the present invention.

For two recognition threshold value groups set in groups, because the similarity of the associated words is higher, but the two recognition threshold value groups are obviously different from other non-associated words, the two recognition threshold value groups are adopted for secondary recognition, whether the associated words belong to a certain associated phrase can be firstly discriminated roughly, the recognition threshold value is improved in the associated phrase, the recognition precision is favorably improved, and the wrong instruction is avoided being sent.

If the second recognition cannot recognize, for example, for "air-conditioning cooling" and "air-conditioning heating", the speaker may pronounce "cold" and "hot" for its own reason, and the device may recognize that the words belong to the relevant phrase, but it is difficult to recognize cooling or heating, and then the device switches to change the state directly according to the current operating state of the device.

Fig. 2 shows a specific implementation of a speech recognition device capable of implementing the speech recognition method, where the speech recognition device includes a command lexicon, a recognition module, and an output module, where the command lexicon includes common words and associated words, the associated words are associated pairwise, pronunciation of the associated words has similarity and the commands are complementary, the speech recognition device further includes a device state detection module, and the recognition module is further connected to the device state detection module through a signal. The identification module may include a primary identification module and a secondary identification module in signal connection with the device status detection module.

The foregoing is a description of preferred embodiments of the present invention, and the preferred embodiments in the preferred embodiments may be combined and combined in any combination, if not obviously contradictory or prerequisite to a certain preferred embodiment, and the specific parameters in the examples and the embodiments are only for the purpose of clearly illustrating the inventor's invention verification process and are not intended to limit the patent protection scope of the present invention, which is defined by the claims and the equivalent structural changes made by the content of the description of the present invention are also included in the protection scope of the present invention.

Claims (2)

1. A speech recognition method, comprising the steps of:

the command words with complementary functions and similar pronunciations are grouped and associated pairwise, wherein the complementary functions refer to that the equipment can only work in a non-A or B working state; and sets first recognition threshold groups a1 and B1, and second recognition threshold groups a2 and B2, respectively, for two command complements within the same group; and a1 is less than a2, B1 is less than B2;

recognizing the command word, judging whether the command word threshold value is larger than any one of the first recognition threshold value group A1 and B1 of any pair of command complementary words, if so, continuously judging whether the command word threshold value is larger than any one of the second recognition threshold value group A2 and B2;

if the current working state of the equipment is not greater than the second recognition threshold set but is greater than any one of the first recognition threshold set, the equipment is switched to the complementary other state;

if the recognition threshold value is not greater than any of the first recognition threshold value groups a1 and B1, the comparison recognition with the other command words is continued.

2. The speech recognition method of claim 1, wherein the first set of recognition thresholds a1 and B1 are set for similar parts in a command word.

Images