Arabic emphatic consonants are primarily cued by coarticulatory effects on adjacent vowels, yet little is known about how these effects unfold across different vowels. This study examined the temporal dynamics of emphatic coarticulation in Saudi Hijazi Arabic across three vowel contexts (/iː/, /aː/, /uː/) and two manners of articulation (stops, fricatives). The plain–emphatic contrast was robust across vowels and manners of articulation for F2 and less consistent for F1. The F2 contrast persisted longest in /aː/, followed by /uː/ and /iː/, with estimated differences also largest in /aː/. Vowel-dependent temporal dynamics further our understanding of emphatic coarticulation beyond static measurements.
Adequately processing amplitude modulation is crucial for speech intelligibility. When faced with concurrent speech streams, background modulations interfere with the perception of the target signal. This study investigated modulation detection interference (MDI) and its relationship to speech perception in noise (SPiN). Results indicated maximal interference when background modulation rates matched that of the target. Interference declined as background modulation rates diverged. However, neither absolute modulation sensitivity nor MDI susceptibility significantly related to SPiN performance. Modulation processing is thus governed by across-channel mechanisms specifically tuned to envelope fluctuations, and does not appear to relate to SPiN directly in adults with typical hearing.
Adults with normal hearing use fundamental frequency and acoustic correlates of vocal tract length (VTL) for gender identification. The goal of this study was to identify the impact of the range of VTL cues on gender judgments. The fundamental frequency and VTL of resynthesized words were independently scaled, and normal hearing participants were assigned to hear stimuli with either a wide or narrow range of VTL values in a gender identification task. These VTL manipulations yielded larger changes in perceived gender in the narrower than the wider condition, highlighting the impact of the range of stimulus parameters on VTL use.
This paper introduces Zimtohrli, an efficient full-reference audio similarity metric. Existing metrics are standardised but require a licensed implementation [Perceptual Objective Listening Quality Assessment (POLQA)], are speech-calibrated [Perceptual Evaluation of Speech Quality, Virtual Speech Quality Objective Listener (ViSQOL)], or—like the audio-domain Perceptual Evaluation of Audio Quality (PEAQ) standard (ITU-R BS.1387)—are tuned to the near-transparent range of BS.1116 tests, leaving a gap for an open metric spanning speech and full-band audio across a wide quality range. Zimtohrli pairs a 128-channel equivalent rectangular bandwidth-spaced gammatone front-end with a non-linear resonator, comparing perceptually-mapped spectrograms via modified dynamic time warping and a neurogram similarity index, optimized against human ratings. It outperforms ViSQOL and Advanced PEAQ, trails POLQA by ∼7% yet surpasses it on full-band music, running 5× faster.
For horizontal arrays in shallow-water waveguides, methods based on the plane wave assumption often produce biased bearing estimates because of multimode propagation. Although subspace-intersection methods reduce bias, they generally require a large array aperture to achieve satisfactory performance. This letter proposes a broadband bearing estimation method based on block sparse Bayesian learning. The proposed method groups modal steering vectors into a block-structured dictionary and exploits joint block sparsity across frequencies to estimate the bearing while mitigating estimation bias. Simulation and experimental results demonstrated that the proposed method provides sharper bearing peaks and better additive noise robustness than the comparison methods.
