Noise-robust speech triage

A method is presented in which conventional speech algorithms are applied, with no modifications,to improve their performance in extremely noisy environments. It has been demonstrated that, foreigen-channel algorithms, pre-training multiple speaker identification (SID) models at a lattice ofsignal-to-noise-ratio (SNR) levels and then performing SID using the appropriate SNR dependentmodel was successful in mitigating noise at all SNR levels. In those tests, it was found that SID performancewas optimized when the SNR of the testing and training data were close or identical. Inthis current effort multiple i-vector algorithms were used, greatly improving both processingthroughput and equal error rate classification accuracy. Using identical approaches in the samenoisy environment, performance of SID, language identification, gender identification, and diarizationwere significantly improved. A critical factor in this improvement is speech activity detection(SAD) that performs reliably in extremely noisy environments, where the speech itself is barelyaudible. To optimize SAD operation at all SNR levels, two algorithms were employed. The firstmaximized detection probability at low levels (10 dB SNR < 10 dB) using just the voicedspeech envelope, and the second exploited features extracted from the original speech to improveoverall accuracy at higher quality levels (SNR10 dB).