Real-time frequency-based noise-robust Automatic Speech Recognition using Multi-Nets Artificial Neural Networks: A multi-views multi-learners approach

摘要: Automatic Speech Recognition (ASR) is a technology for identifying uttered word(s) represented as an acoustic signal. However, one of the important aspects noise-robust ASR system its ability to recognise speech accurately in noisy conditions. This paper studies applications Multi-Nets Artificial Neural Networks (M-N ANNs), realisation multiple-views multiple-learners approach, Multi-Networks Recognisers (M-NSRs) providing real-time, frequency-based model. M-NSRs define features associated with each word different view and apply standalone ANN learners approximate that view; meanwhile, single-learner (MVSL) ANN-based recognisers employ only memorise entire vocabulary. In this research, M-NSR was provided evaluated using unforeseen test data were affected by white, brown, pink noises; more specifically, 27 experiments conducted on measure accuracy recognition rate proposed Furthermore, results compared detail MVSL system. The recorded improved average up 20.14% when it given infected noise our experiments. It shown higher degree generalisability can handle because has than previous model under