Automated Speech Audiometry for Integrated Voice Over Internet Protocol Communication Services

Area of Science:

• Audiology and Speech Sciences
• Telecommunications Engineering
• Acoustics and Signal Processing

Background:

• Speech recognition challenges are common in telecommunication environments.
• Ecologically valid assessments require quantifying real-world transmission degradations.
• Traditional laboratory speech audiometry may not fully represent real-world communication scenarios.

Purpose of the Study:

• To present an automated speech recognition (ASR) based matrix sentence test procedure integrated into a Voice over Internet Protocol (VoIP) infrastructure.
• To quantify the impact of individual transmission degradations on speech recognition thresholds (SRTs).
• To compare automated VoIP-based SRT measurements with traditional laboratory measurements.

Main Methods:

• Speech recognition thresholds (SRTs) were measured in 16 normal-hearing subjects across four conditions.
• Conditions included a standard laboratory setting, laboratory simulation of reduced bandwidth and headset quality, and an automated ASR-controlled adaptive test over VoIP.
• ASR system errors were analyzed to assess their impact on measurement outcomes.

Main Results:

• A highly significant correlation (r = .93) was found between automated VoIP and laboratory SRT measurements.
• A consistent bias of approximately 1 dB was observed, indicating a linear shift without altering data distribution.
• The individual effects of system degradations on SRTs were successfully quantified.

Conclusions:

• Automated ASR-based SRT measurements via VoIP systems are a feasible proof of concept for speech audiometry.
• Results from the automated VoIP method closely align with traditional laboratory findings for normal-hearing individuals.
• VoIP services present a promising avenue for conducting speech audiometric testing in realistic communication environments.