Neural Network Based Speech Identification/Transcription Model

We propose to adapt our proprietary voice authentication system to provide speaker identification functionality. While the authentication system looks for changes in speaker identity after a normal pause in speech, the identificatin system will compare features of speakers' voices with featurs stored for those speakers in a database. The features will be determined using advanced signal processing techniques, and speaker identification will be performed using advanced statistical analysis of the extracted features, including neural networks. Both of these approaches build on our experience with voice authentication. The system will be designed to function in a noisy, multi-speaker environment, and to interface with speech recognition software.