Efficient Adaptive Speech Reception Threshold Measurements Using Stochastic Approximation Algorithms

Area of Science:

• Audiology
• Speech processing
• Signal processing

Background:

• Assessing speech reception threshold in noise (SRT50n) is crucial for hearing aid and cochlear implant (CI) users.
• Current adaptive procedures for SRT50n estimation may not be optimal, particularly for CI users.
• Stochastic approximation (SA) offers a potential method for improving adaptive testing efficiency.

Purpose of the Study:

• To investigate the optimization of speech-in-noise tests using stochastic approximation (SA) methods.
• To compare the efficiency and accuracy of SA algorithms against traditional adaptive procedures for SRT50n estimation in cochlear implant (CI) users and normal-hearing (NH) listeners.
• To validate a simulation model for assessing SRT50n estimation methods.

Main Methods:

• Development of a simulation model to predict intelligibility scores in noise for CI and NH listeners.
• Monte Carlo simulations were used to compare four optimized SA algorithms with clinically used adaptive procedures.
• The simulation model's validity was confirmed by comparing its results with existing experimental data.

Main Results:

• The simulation model accurately reflected real-world data.
• Four optimized SA algorithms efficiently estimated SRT50n, showing equal accuracy and smaller standard deviations (SDs) compared to clinical methods.
• In CI users, SRT50n estimates exhibited a small bias and larger SDs than in NH listeners, with reliability concerns if speech intelligibility in quiet fell below 70%.

Conclusions:

• Stochastic approximation (SA) algorithms are effective for adaptive speech-in-noise tests across diverse listener groups, including CI and NH individuals.
• SA methods provide efficient SRT50n estimation in CI users, provided their speech intelligibility in quiet exceeds 70%.
• SA procedures represent a valid, more efficient alternative to current clinical adaptive methods for CI users.