Soft computation based spectral and temporal models of linguistically motivated Assamese telephonic conversation recognition

Abstract

Speech is the natural communication means, though not the typical input means afforded by computers. The interaction between humans and machines would have become easier, if speech were an alternative effective input means supplementing the keyboard and mouse. With advancement in techniques for signal processing and model building leading to the empowerment of computing devices with expanding list of abilities, significant progress has been made in speech recognition research, and various speech based applications have been developed. In such a backdrop, telephone speech technology have been receiving more attention in many new applications of spoken language processing. From the literature it has been found that the spectro-temporal features gives a significant performance improvement for telephone speech recognition in comparison to the conventionally used features for speech/speaker identification. Speech recognition systems can be characterized by many parameters. The commonly used method to measure the performance of a speech recognition system is the recognition accuracy. For obtaining proper accuracy it is necessary to design an efficient classifier for the recognition purpose which will lead to correct recognition results.