Acoustic Disturbance Sensing Level Detection for ASD Diagnosis and Intelligibility Enhancement

TL;DR

This work targets speech intelligibility for individuals with Autism Spectrum Disorder under noisy urban conditions. It proposes ISE$_{\text{ASD}}$, a three-stage pipeline that estimates F0 via HHT-Amp, uses a harmonics-centered Gammatone filterbank, and applies gains to emphasize harmonic components, aiming to mitigate external noise masking in ASD. Objective evaluations (ESTOI and PESQ) show ISE$_{\text{ASD}}$ outperforms three baselines across four noises and multiple SNRs, with perceptual tests corroborating intelligibility gains for both NT and ASD groups. The approach also highlights the potential of the harmonic-sensing strategy as an auxiliary aid for ASD diagnosis, while noting computational considerations and avenues for further refinement.

Abstract

The acoustic sensitivity of Autism Spectrum Disorder (ASD) individuals highly impacts their intelligibility in noisy urban environments. In this Letter, the disturbance sensing level is examined with perceptual listening tests that demonstrate the impact of their append High Internal Noise (HIN) profile on intelligibility. This particular sensing level is then proposed as additional aid to ASD diagnosis. In this Letter, a novel intelligibility enhancement scheme is also introduced for ASD particular circumstances. For this proposal, harmonic features estimated from speech signal frames are considered as center frequencies of auditory filterbanks. A gain factor is further applied to the output of the filtered samples. The experimental results demonstrate that the proposal improved the acoustic intelligibility of ASD and Neurotypicals (NT) people considering four acoustic noises at different signal-to-noise ratios.

Figures (6)

• Figure 1: (a) Illustration of the High Internal Noise impact in the acoustic intelligibility of ASD individuals.
• Figure 2: Perceptual Threshold versus Noise (TvN) for ASD versus NT individuals PARK_2017.
• Figure 3: Block diagram of the proposed ISE$_\text{ASD}$ method for speech intelligibility gain for ASD individuals.
• Figure 4: ESTOI average scores to characterize the personalized gains for speech signals corrupted by SSN noise with SNR values: (a) -5 dB, (b) 0 dB and (c) 5 dB.
• Figure 5: ESTOI improvement [$\times10^{-2}$] boxplots for the noisy speech signals in different SNR values processed with the competitive methods: SSFV, APES$_\text{HARM}$, GTF$_\text{F0}$ and ISE$_\text{ASD}$.