Measuring Mandarin Speech Recognition Thresholds Using the Method of Adaptive Tracking.

Abstract

Purpose The purpose of this study was to examine (a) the psychometric functions of Mandarin speech-on-speech masking with a comparison to the psychometric properties of the English version and (b) the feasibility of using the method of adaptive tracking (MAT) to measure Mandarin speech recognition thresholds in different masker conditions using a Chinese coordinate response measure (C-CRM) speech corpus. Method The study followed the example of the English coordinate response measure (E-CRM) speech corpus and designed a Chinese version. Sixteen listeners were recruited for a monaural Mandarin speech recognition test using C-CRM materials in 5 competing speech masker conditions: 1-talker different-gender, 1-talker same-gender, and 2-, 4-, and 8-talker maskers. Both the MAT and the method of constant stimuli (MCS) were used and then compared. Results Results of the MCS revealed that Mandarin listeners psychometric functions varied depending upon masker characteristics (e.g., the number and gender of talkers), because significant differences were suggested among the 5 masker conditions (mainly between 1-talker and multitalker), as measured in the slope b and the 50% threshold. Meanwhile, speech identification improved monotonically with increasing signal-to-noise ratios for all 5 masker conditions, except that a plateau was observed with signal-to-noise ratios from -10 to 0 dB in the 1-talker maskers. The properties of C-CRM psychometric functions largely agreed with E-CRM findings, and yet lower speech recognition thresholds were found in C-CRM. Meanwhile, the comparisons of the thresholds estimated in the MAT and the MCS showed no significant differences for all 5 masker conditions. Conclusion The results suggest that masker characteristics greatly affected Mandarin psychometric functions and monotonicity was present in all 5 C-CRM maskers, echoing the E-CRM literature in a great extent. Interestingly, target words of C-CRM were found easier to recognize in multitalker babbles than those of E-CRM. Moreover, comparisons of the MAT and the MCS suggest that the MAT provided a feasible method for measuring Mandarin speech recognition thresholds with the C-CRM speech maskers.