The research investigates how consonants impact word intelligibility, a critical factor in motor speech disorder intervention. The study proposes a method to measure this contribution by systematically masking individual consonants within isolated words. An automatic speech recognition (ASR) model is used to determine if the word remains recognizable after the consonant is silenced. The resulting ‘mask-induced misrecognition rate’ (MMR) serves as a metric for quantifying a consonant’s contribution. This approach offers a scalable alternative to traditional perceptual studies.
The research utilized three ASR architectures: MMS (encoder-only), Whisper (encoder-decoder), and Qwen3-ASR (LLM-based). Analysis of the MMR across four languages – English, Spanish, German, and Czech – revealed that phoneme frequency correlates negatively with MMR, indicating that more frequent consonants are less disruptive when masked. Conversely, functional load correlates positively with MMR, suggesting that consonants carrying more lexical contrast are more disruptive.
Cross-language analysis demonstrated that consonant rankings are not consistent across languages. This finding highlights the language-dependent nature of consonant contribution. The study’s results provide a framework for prioritizing intervention targets in motor speech disorders by focusing on consonants with higher MMR values. The method’s scalability allows for efficient analysis across diverse languages and ASR models.
The research utilized partial Spearman correlations to assess the relationships between linguistic factors and MMR. The findings contribute to a better understanding of the acoustic properties of speech and their impact on comprehension. Further investigation could explore the impact of other acoustic features and speech production parameters.
Source: https://arxiv.org/abs/2609.12122