Voice-pulse conversion method based on the response of the primary auditory cortex

Brain machine interface (BMI) is one of useful techniques to compensate the deteriorated sensory functions. For the purpose of aid of auditory impediment, we have developed a sound-stimulus signal conversion interface for an auditory BMI that is based on the response of primary auditory cortex (A1) cells. In A1 neuron responses, amplitude-modulated (AM) sound is encoded as spike rate information of the tonic response, and frequency-modulated (FM) sound is encoded as the order of active place in the brain. In signal processing, the sound signals were divided into frequency bands by using a digital filter bank for FM sound. The separate signals were then converted to pulse train signals to deal with AM. In this manner, a voice-pulse conversion method that is based on A1 neuron response to AM and FM sound was constructed. Additionally, in order to examine which frequency band division method was suitable for vowel identification, we got the self-organizing map to learn the number of pulses for each output channel. The result showed that the exponential processing was best. Under the condition, the formant information is left behind and the classification of vowels is possible.