Speech-in-Speech Recognition: Effect of Language Uncertainty

Abstract

Speech-in-speech recognition is a common phenomenon, in a multilingual country like India. This phenomenon co- exists with unpredictable change in the background language, especially with English language. The present study aims to study the influence of native vs. non-native language babble, language uncertainty of the babble and role of signal-to-noise ratio on the Kannada sentence recognition. Forty Kannada-English Bilinguals were subjected to Kannada sentence recognition task in the presence of two-talker Kannada babble, two-talker English babble, two- talker mixed (Kannada + English) babble and two-talker interleaved (unpredictable change in background language) at +3 dB SNR, 0 dB SNR and -3 dB SNR. The level of presentation of target sentence was 75 dB SPL. Comparison of the recognition scores across four babble conditions revealed significantly better scores in the presence of two-talker English babble when compared to two-talker Kannada babble, implying the release of masking in the presence of non-native language babble. Further, the presence of language uncertainty in the two- talker interleaved condition reduced the scores when compared to two-talker mixed condition. The influence of other variables like F0 and LTASS differences influencing the speech-in-speech recognition was negligible and the release of masking can be attributed to the linguistic mismatch of the target and the babble. In addition, SNR was found to be a factor influencing speech-in-speech recognition.