Skip to main content

Core Software

HAAQI Speech Intelligibility model

This is a python implementation of the Hearing Aid Audio Quality Index (HAAQI) model which is used for objective estimation. This will be used in the stage 1 evaluation of entrants (see Rules).

Note that HAAQI is not a binaural metric, instead, each channel must be processed separately. We average the left and right scores to produce a final overall score.

Instructions and recommendations on the use of HAAQI

You can call the HAAQI metric using the clarity.evaluator.haaqi.compute_haaqi function as:

  1. The processed_signal corresponds to the output signal from enhancement block in the baseline.
  2. The reference_signal corresponds to the amplified reference signal. This is generated by the evaluation block in the baseline.
  3. It is recommended to resample the processed_signal and reference_signal signals to 24,000 Hz before calling the metric.
  4. Set the parameters processed_sample_rate and reference_sample_rate equal to 24000.
  5. The audiogram corresponds to the left or right audiogram object.
  6. Set equalisation to 2, which indicates to HAAQI that the reference_signal has NAL-R applied to it.
  7. Set the level1 parameter to the level of the reference signal before applying the hearing aid amplification (NAL-R). Recommended to set it as 65 - 20 * log10(RMS(reference signal before NAL-R))

Please note that the level1 parameter uses the reference signal without NAL-R amplification. And, the reference_signal expects the amplified version of the same signal used in level1.

4. References

[1] Byrne, Denis, and Harvey Dillon. "The National Acoustic Laboratories'(NAL) new procedure for selecting the gain and frequency response of a hearing aid." Ear and hearing 7.4 (1986): 257-265.