lars_round
Lars Bramsløw

Research Engineer, PhD, Project Leader

> Learn more

mail@eriksholm.com

Speech Reception Threshold is often used to measure differences between alternative hearing-aid systems. But when the dB range of the SRT measurement is not constrained to keep it relevant for the systems and listeners involved, the results may be invalid or misleading. We suggest ways to solve this problem.

Background

SRT-like measurements for quantifying the intelligibility of speech in noise, with results in dB SNR, were originally developed as efficient diagnostic tools for evaluating speech-reception disability at the start of a clinical treatment process. Nowadays it is increasingly common for research into hearing-aid systems to utilise SRT measurements when quantifying outcomes; however, the validity and relevance of such approaches is open to question - especially with the increasing interest in highly non-linear signal processing schemes in hearing aids.

SRT is an attractive outcome measure, as it can be made relatively fast and repeatable, and ‘always yields a result’ with convenient statistical properties -and no floor or ceiling effects. However, the first priority for outcome measures in HA research should be validity and relevance rather than convenience.

Analysis

Various factors influence the suitability of SRT as an outcome measure. The fundamental problem is that SRT procedures do not adequately constrain the Signal-Noise Ratio (SNR) at which testing takes place. The behaviour of non-linear systems, even simple ones like dynamic range compression, is highly dependent on the SNR at the input; thus the results of comparisons between alternative non-linear (i.e. almost any) hearing-aid systems may be critically affected by:
 
• Aspects of the SRT procedure itself - corpus, scoring, noise type, acoustics etc., which affect the SNR at which criterion %-correct is achieved

• The inherent SNR needs of the listener, which depend on audiometric loss, cognitive abilities, etc.
 
Examining past studies on the influence of such factors on SRT, the aggregated range of variation in achievable SRTs is apparently absurdly large. We must therefore conclude that the effects of the various factors are not independent, but subject to, possibly strong, interactions. The issue of ecological validity must also be considered; testing should preferably take place in an SNR range relevant for the user (i.e. realistic), and for the signal processing features under test. In order to address ecological validity, robust documentation of real-life SNRs to provide meaningful constraints on test SNRs is greatly in demand.

Experiments with ‘SRT manipulators’

The procedural aspects of SRT measurements mentioned above can also be used constructively to shift the SRT of an individual listener towards a desired value. Such SRT manipulators were investigated in two studies with hearing-impaired listeners using their own hearing aids. Target sentences were presented from the 0º loudspeaker seen in Figure 1, while various configurations of background talkers were presented from the other loudspeakers. For example, SRTs were determined for each listener with background talkers from either ±15º, ±30º, ±35º, or ±75º, with either male or female background talkers, and with either 50% correct words or 50% correct sentences as the criterion.

  • Limitations of Speech Recognition Thresshold Fig 1
    Figure 1. Loudspeaker set-up used for the SRT manipulator experiments in Eriksholm’s anechoic chamber.

Results

Useful SRT shifts were found for all three SRT manipulators mentioned above - target-to-masker angular separation, background talker gender, scoring method. As an example, the effect of changing the spatial separation between target and background talkers is shown in Figure 2.

When using the Danish HINT sentences as target, the full range of possible SRT manipulation was 8 dB on average, which corresponds well with the variation between listeners in any given condition. These manipulators were used in the development and validation of the Spatial Fixed SNR (SFS) test which allows different members of a group of hearing-impaired listeners to be tested at the same SNR, including some control of the ecological validity of the test SNR.

  • Limitations of Speech Recognition Thresshold Fig 2
    Figure 2. Mean SRT across 20 listeners for different target-background separation angles.

Learn more

  • learnmore-poster

    Presentation

    Limitations of speech reception threshold (SRT) as an outcome measure

  • learnmore-poster

    Poster

    Towards a spatial speech-in-speech test that takes SNR confounds and ecological validity into account

  • learnmore-poster

    Poster

    Investigation of SRT 'manipulators' for a spatial speech-in-speech test that takes SNR confounds and ecological validity into account

  • learnmore-poster

    Poster

    Investigating SRT 'manipulators' for a spatial speech-in-speech test
  • learnmore-poster

    Poster

    Magnitude of SRT manipulators
  • learnmore-poster

    Poster

    Validation of a spatial speech-in-speech test that takes signal-to-noise ratio (SNR) confounds into account