KR20010088766A - Speech recognition control system using dsp and wireless communication - Google Patents

Abstract

PURPOSE: A speech recognition control device using a DSP(Digital Signal Processing Unit) and radio communication is provided to recognize input speech signals through a speech recognition process and transfer the recognized result to a control device through radio communication, thereby solving a problem of error recognition. CONSTITUTION: A speech input part receives speech signals. A DSP implements a speech recognition algorithm to recognize input speech signals. A radio communication part transfers the recognized speech signals through a radio unit. A control device controls a speech output part on the basis of the recognized speech signals to inform the state of a controller, or controls the controller to realize desired operations.

Description

SPEECH RECOGNITION CONTROL SYSTEM USING DSP AND WIRELESS COMMUNICATION}

The present invention applies voice recognition to a first control device, and more particularly, to a voice recognition control device using a DSP and wireless communication.

The device can improve the quality of life by, for example, automating the functions of various facilities in the home, thereby increasing the landlord's ability to control the housing space. In the case of home automation, an example of a control device, the method using a remote controller is mainstream.

However, the use of such a remote control can be a significant inconvenience for the disabled. That is, such a remote control must be operated by using a hand. However, a handicapped person who has an uncomfortable hand cannot operate a remote control, and thus, home automation cannot be used.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and an object of the present invention is to provide a voice recognition control device using a DSP and wireless communication, which enables a disabled person to use a control device.

1 is an overall system configuration of the voice recognition control device,

2 is a view showing a recognition rate of the voice recognition control device.

In order to solve the above problems, the present invention constitutes a voice recognition control system using DSP and RF communication.

The voice recognition control device is largely composed of a voice input unit, a digital signal processing unit (DSP), a wireless communication unit (RF module), and a control system unit (Control System).

LPF (Lowpass Filter) is used as a preprocessor to remove noise in the voice input unit, and an A / D converter having an 8-bit resolution for converting an analog signal into a digital signal Analog / Digital converter) AD7821KN is used.

The signal passing through the A / D converter performs the voice recognition process using the DSP unit that has a high operation speed, and wirelessly transmits the recognized word to the controller unit through the RF module. In the control unit, MICOM uses an 8-bit microprocessor called 80c51 and uses ROM (ROM, 27256) and RAM (RAM, 62256) as external memory. The MICOM is configured to identify the signal received from the RF module and inform the current state of the recognized word by voice and to drive the control unit as the command recognized as 'yes' or 'no'. The chip used for voice output is the ISD2560, which is a very simple chip for recording and playing back voice.

(Example)

Hereinafter, the overall configuration of the present invention will be described with reference to the drawings taking home automation as an example.

First, the voice input unit will be described.

Real time processing is important when constructing an independent voice recognition system. In particular, in order to find and process a section in which speech is actually contained, a calculation must be performed every time one or several speech samples are received. To do this, the system performs an external interrupt every time one sample comes in to receive the data and exits to perform the calculation process in order to perform the calculation process from one voice sample to the next. Organize circuits and programs to enable

The A / D converter is configured to have a sampling frequency of 8 kHz and a quantization rate of 8 bits of resolution, and adds an LPF with a 4 kHz cut-off frequency.

Next, the DSP unit and the voice recognition process.

The DSP used in the present invention is a TMS320C3X series, which is a 32-bit floating point method released by Texas Instruments. It has the function of hardware processing high speed computing circuits such as multipliers, so it can process large amounts of data or complex operations in a short time. The functions and features of the DSP used are as follows.

Once the voice comes in, the process of searching for the section with the actual voice from the input voice should be executed. This is called detection of a real voice interval. In speech recognition, accurate real speech section detection is a very important process along with the recognition algorithm. In the present invention, the detection of the real speech section is accurately performed by using a function that simultaneously uses the energy and the boundary crossing rate for the detection of the real speech section.

If a real speech section is detected, feature parameters must be extracted from the real speech. From this feature parameter, recognition is performed through a process of learning and comparison. Extraction of speech feature parameters is very important to effectively describe speech with as little numerical data as possible.

In the present invention, the cepstrum coefficient and the log scale energy derived from the 10-dimensional LPC coefficients are used as the 11th order parameters.

After extracting feature parameters for each command, vector quantization (VQ) is performed to use them in the codebook and training process. VQ was developed with the intention of increasing the transmission rate by reducing the information rate of the voice signal.

The VQ process can be divided into two types: a codebook generation process and an observation sequence generation process.

First, a codebook is generated through the VQ process, and an observation sequence is generated from the coefficients to be trained. The method of generating the observation sequence compares the distance between the coefficients to be trained and the reference pattern, and then uses the index of the vector closest to the distance. When the vocabulary for recognition comes in, the training sequence is generated again in comparison with the codebook generated from the VQ process as in the training.

Speech training and recognition algorithms use the Hidden Macov Model, which can accurately reflect the time-varying nature of speech signals.

The vocabulary used for recognition is / TV /, / video /, / air conditioner /, / heating /, / entrance /, / computer /, / washing machine /, / memo /, / cooking /, It consisted of 10 words such as / gas / and trained with Baum-Welchi algorithm for 10 models. In order to set up each model, the VQ level consists of 64. It is assumed that the phoneme has no initial syllable 'ㅇ' of syllable.

When the test vocabulary comes in, as in training, the observation sequence is generated by comparing the input pattern with the reference pattern previously generated from the VQ process. This observation string and each trained probability value are applied to the Viterbi algorithm. The Viterbi algorithm finds the optimal state sequence for a given observation sequence and model. The probability value is calculated from the determined optimal state string. It is recognized as the model with the highest probability among the probability values calculated for each model.

Topology applies a model that can transition only one state of the left-to-right model. Each model is trained according to the number of states you set.

In the training method, first, a model is set, and the pi, A, and B values of the set model are initialized. Where π is the initial state probability distribution, A is the state transition probability, and B is the observed symbol probability in state j. At this time, the initial values of A and B were all random. Always set the value of the first state to 1 and the rest to 0 so that it always transitions from the first state.

This recognition process is performed for each model, and the model with the highest probability among them becomes the recognition result.

The recognized result is wirelessly transmitted to the control unit through the RF module connected to the DSP unit. The controller unit voices the current state of the recognized word and automatically executes it. This execution method is as follows. First of all, the controller informs the status of the recognized word. (Example: TV is on. Are you sure you want to turn it off?) If you answer yes or no to a given voice question, the corresponding command will be performed.

As described above, the present invention drives the voice recognition control device by recognizing the input voice signal in the voice recognition process and sending the result to the control device using wireless communication. The recognition rate of the proposed speech recognition control device is shown in FIG.

As can be seen in the figure, the recognition rate for each word can be seen that the average of more than 90% independent of the speaker. In addition, the controller unit informed the status of the recognized words (eg, the air conditioner is turned off. Do you want to turn it on?) To solve the problem of false recognition, which is an important problem in speech recognition. Therefore, this invention can acknowledge the utility.

Advantages of the present invention include simplicity, safety, high quality, entertainment, accuracy, savings and the like. As a part which can be applied to the present invention, indoor and outdoor lighting systems, security facilities, heating and cooling facilities, door and garage door opening and closing facilities, audio / video facilities, kitchen facilities and the like can be listed.

Claims (1)

A voice input unit for receiving a voice signal, A DSP unit for recognizing an input voice signal by implementing a voice recognition algorithm; A wireless communication unit for shearing the recognized signal to the control unit through radio; Based on the recognized signal, it controls the voice output unit to inform the status of the control device or the control device to control the control device to perform the desired operation. Voice recognition control device using DSP and wireless communication.