51单片机语音识别外文文献及翻译
Implementation of Speech Recognition on MCS51 Microcontroller for Controlling Wheelchair
Thiang
Electrical Engineering Department, Petra Christian University
I.INTRODUCTION
Automatic speech recognition by machine has been a goal of research for more than four decades. However, in spite of the glamour of designing an intelligent machine that can recognize the spoken word and comprehend its meaning, and in spite of enormous research efforts spent in trying to create such a machine, it is far from achieving the desired goal of a machine that can understand spoken discourse on any subject by all speakers in all environments. The speech recognition system has also been implemented on some particular devices. Some of them are personal computer (PC), digital signal processor, and another kind of single chip integrated circuit. This paper introduces the speech recognition which was implemented on a microcontroller. The microcontroller where the speech recognition was implemented on is ATMEL AT89C51RC. This microcontroller is a MCS51 family microcontroller and this microcontroller was chosen because it is popular in Indonesia. The speech recognition system that is implemented on the microcontroller is used to recognize the word in a speech signal. The words are used as the command for controlling movement of a wheelchair. Therefore, the system was designed to recognize limited number of the words. This is also caused by the limit number of data memory of the microcontroller. There are only seven words used as the command for controlling movement of the wheelchair. They are stop, forward, backward, left, right, up, and down which is used to stop the wheelchair, to move forward the wheelchair, to move backward the wheelchair, to turn left the wheelchair, to turn right the wheelchair, to increase speed of th优/文^论'文.网http://www.youerw.com e wheelchair, and to decrease speed of the wheelchair respectively. Two approaches were implemented to perform the speech recognition. The first approach is Linear Predictive Coding (LPC) and Euclidean Squared Distance (ESD). LPC is used as the feature extraction method and ESD is used as the recognition method. This approach is based on the pattern recognition approach. The second approach applied in this system is Hidden Markov Model (HMM), which is one of the speech recognition approaches that classified as the statistical pattern recognition. HMM is used as the recognition method. As the feature extraction method, a simple segmentation and centroid value is applied. Section 3 and 4 of this paper describe about this two approaches more detail. The mechanism and hardware design of the wheelchair is explained in the section 2 of this paper.
II. HARDWARD DESIGN OF WHEELCHAIR
Figure 1 shows block diagram of hardware system.
Hardware of the system consists of three main parts. The first part is DC motor control circuit. The circuit consists of controller, driver, and DC motor speed sensor circuit. In this part, an ATMEL AT89C52 microcontroller is used as the controller. The second part is microcontroller minimum system which performs the speech recognition process and microphone interface. In the second part, an ATMEL AT89C51RC microcontroller is used as the speech recognition processor. The third part is interface circuit. This circuit performs the communication between DC motor controller and speech recognition processor. This circuit also read the input command from a set of keypads. An ATMEL AT89C2051 microcontroller is used as the interface in this part. The speech recognition system was implemented on ATMEL AT89C51RC microcontroller which runs at frequency of 24 MHz and has 32k bytes program memory. With a 24 MHz clock, the fastest time to execute an instruction by the microcontroller is about 0.5 microseconds.
Fig 1. Block Diagram of Hardware System
III. SPEECH RECOGNITION USING LINEAR PREDICTIVE CODING AND EUCLIDEAN SQUARED DISTANCE
The first approach of speech recognition implemented on microcontroller is Linear Predictive Coding (LPC), which is combined with Euclidean Squared Distance (ESD) method. LPC is used as the feature extraction method and Euclidean Squared Distance is used as the recognition method. Blockdiagram of LPC and ESD training and recognizer system are shown at figure 2 and 3 respectively .In training system, training data are sampled directly from microphone. Then, each training sample is processed using LPC processor algorithm and the result of this process is a set of cepstral coefficients of the speech signal. These cepstral coefficients are used as the reference model. A simple algorithm was implemented to detect the existence of the speech signal. The system reads four consecutive sampling data and then calculates the average of those four data. If the average value is less than a limit value, it means there is no speech signal. If the average value is greater than or equal to that limit value, it means there is a speech signal and then the microcontroller will start to read and record the signal in 0.5 seconds.
Fig 2. Block Diagram of Training System Using LPC
Fig 3. Block Diagram of Recognizer System Using LPC and ESD
IV. SPEECH RECOGNITION USING HIDDEN MARKOVMODEL
The second approach of speech recognition implemented on microcontroller is Hidden Markov Model (HMM) which is used as the recognition method. In this approach, LPC processor is not used as the feature extraction method because calculation of LPC processor takes much time (about 19 seconds) when it was implemented on AT89C51RC. Therefore, instead of the LPC processor, a simple feature extraction algorithm, segmentation and centroid, was implemented to reduce the calculation time.1621