Enhancing listener strategies using a payoff matrix in speech-on-speech masking experiments

Area of Science:

• Auditory perception
• Speech processing
• Psychoacoustics

Background:

• Understanding speech recognition in noisy environments is crucial.
• Speech maskers, especially those syntactically similar, pose significant challenges.
• Listener behavior can influence speech recognition performance.

Purpose of the Study:

• To investigate speech recognition performance as a function of target-to-masker ratio (TMR).
• To evaluate the impact of listener instructions and feedback on speech recognition with similar speech maskers.
• To determine optimal listening conditions for speech intelligibility.

Main Methods:

• Speech recognition was measured using varying target-to-masker ratios (TMR).
• Two experiments were conducted with syntactically similar speech maskers from the same talker.
• Listener instructions and a payoff matrix were manipulated to influence reporting behavior.

Main Results:

• Performance plateaued below 0 dB TMR in the first experiment, with varied listener strategies.
• In the second experiment, listeners reduced masker keyword reports and improved target keyword reports.
• A local minimum in performance was observed at 0 dB TMR with same-talker maskers.
• Optimal performance was achieved with a 9 dB level difference, favoring the target speaker.

Conclusions:

• Listener instructions and feedback significantly impact speech recognition in the presence of similar speech maskers.
• A level difference favoring the target speaker enhances intelligibility.
• Understanding and mitigating listener bias is key for improving speech recognition technology.