Phoneme recognition system.
James Lisonbee, Dan Tebbs, Paul A. Wheeler · The Journal of the Acoustical Society of America · 1992
The possibility of phoneme recognition using a hardware-based spectral sampling filter and a 386-μ processor software-based neural network to accurately recognize ten spoken vowel sounds of the English language was explored. To apply the neural network to speech recognition, a set of inputs conducive to the neural network architecture must be supplied. The spectral sampling filter does this by creating a pattern representing energy content in the first and second formants. This analog pattern is converted to digital and communicated to the neural network. The neural network was trained using a back-propagation technique. The neural network sees the localized energy changes across time as a distinct pattern for each phoneme. The neural network executes a recognition sequence and produces an integer correlating to the spoken phoneme.