A radio controller using speech for the blind

This article describes a low-cost, portable real-time DSP-based speech controller system to provide radio interface control command applications for the blind. The system recognizes spoken Mandarin Chinese words on a DSP chip (TMS320C31) using a hidden Markov model. The function of the radio set, which includes a tuner, tape, and compact disc, were evaluated under both noisy and noiseless environments. Four subjects took part in the experiment and achieved 83 and 90% mean recognition rates under noisy and noiseless conditions, respectively. In addition, because this system is based on a DSP chip, it can easily be programmed to execute speaker-independent algorithms.

Isokinetic elbow joint torques estimation from surface EMG and joint kinematic data: using an artificial neural network model

Because the relations between electromyographic signal (EMG) and anisometric joint torque remain unpredictable, the aim of this study was to determine the relations between the EMG activity and the isokinetic elbow joint torque via an artificial neural network (ANN) model. This 3-layer feed-forward network was constructed using an error back-propagation algorithm with an adaptive learning rate. The experimental validation was achieved by rectified, low-pass filtered EMG signals from the representative muscles, joint angle and joint angular velocity and measured torque. Learning with a limited set of examples allowed accurate prediction of isokinetic joint torque from novel EMG activities, joint position, joint angular velocity. Sensitivity analysis of the hidden node numbers during the learning and testing phases demonstrated that the choice of numbers of hidden node was not critical except at extreme values of those parameters. Model predictions were well correlated with the experimental data (the mean root-mean-square-difference and correlation coefficient gamma in learning were 0.0290 and 0.998, respectively, and in three different speed testings were 0.1413 and 0.900, respectively). These results suggested that an ANN model can represent the relations between EMG and joint torque/moment in human isokinetic movements. The effect of different adjacent electrode sites was also evaluated and showed the location of electrodes was very important to produce errors in the ANN model.

A neuro-control system for the knee joint position control with quadriceps stimulation

A neuro-control system was designed to control the knee joint to move in accordance with the desired trajectory of movement through stimulation of quadriceps muscle. This control system consisted of a neural controller and a fixed parameter proportional-integral-derivative (PID) feedback controller, which was designated as a neuro-PID controller. A multilayer feedforward time-delay neural network was used and trained as an inverse model of the functional electrical stimulation (FES)-induced quadriceps-lower leg system for direct feedforward control. The training signals for neural network learning were obtained from experimentation using a low-pass filtered random sequence to reveal the plant characteristics. The Nguyen-Widrow method was used to initialize the neural connection weights. The conjugate gradient descent algorithm was then used to modify these connection weights so as to minimize the errors between the desired outputs and the network outputs. The knee joint angle was controlled with only small deviations along the desired trajectory with the aid of the neural controller. In addition, the PID feedback controller was utilized to compensate for the residual tracking errors caused by disturbances and modeling errors. This control strategy was evaluated on one able-bodied and one paraplegic subject. The neuro-PID controller showed promise as a position controller of knee joint angle with quadriceps stimulation.

Real-time implementation of electromyogram pattern recognition as a control command of man-machine interface

The purpose of this study was to develop a real-time electromyogram (EMG) discrimination system to provide control commands for man-machine interface applications. A host computer with a plug-in data acquisition and processing board containing a TMS320 C31 floating-point digital signal processor was used to attain real-time EMG classification. Two-channel EMG signals were collected by two pairs of surface electrodes located bilaterally between the sternocleidomastoid and the upper trapezius. Five motions of the neck and shoulders were discriminated for each subject. The zero-crossing rate was employed to detect the onset of muscle contraction. The cepstral coefficients, derived from autoregressive coefficients and estimated by a recursive least square algorithm, were used as the recognition features. These features were then discriminated using a modified maximum likelihood distance classifier. The total response time of this EMG discrimination system was achieved about within 0.17 s. Four able bodied and two C5/6 quadriplegic subjects took part in the experiment, and achieved 95% mean recognition rate in discrimination between the five specific motions. The response time and the reliability of recognition indicate that this system has the potential to discriminate body motions for man-machine interface applications.