Does Sentence-Level Coarticulation Affect Speech Recognition in Noise or a Speech Masker?

Area of Science:

• Auditory Perception
• Speech Processing
• Psychoacoustics

Background:

• Between-word coarticulation, the influence of adjacent speech sounds, is a natural aspect of spoken language.
• Understanding coarticulation's role in speech recognition is crucial for developing effective auditory aids and speech processing technologies.

Purpose of the Study:

• To investigate the impact of natural between-word coarticulation on masked speech recognition.
• To compare speech recognition in naturally coarticulated sentences versus sentences constructed from individually spoken words.
• To test hypotheses regarding coarticulation benefits in different masking conditions (noise vs. speech).

Main Methods:

• Sixty young adults participated in three experiments using an adaptive tracking procedure to determine sentence recognition thresholds (SRTs).
• Stimuli included naturally coarticulated sentences, concatenated sentences, and concatenated keyword-intensity-matched (KIM) sentences.
• Experiments utilized both steady noise and two-talker speech maskers.

Main Results:

• Overall effects of coarticulation on SRTs were small, generally less than 2 dB.
• Coarticulation showed a benefit compared to concatenated KIM sentences in one experiment.
• A coarticulation benefit was observed in noise when semantic context was available, but not in a speech masker.

Conclusions:

• Differences in speech recognition between coarticulated and non-coarticulated sentences were minimal.
• The audibility of concatenated, unscaled sentences may offset potential coarticulation benefits.
• Contrary to hypotheses, coarticulation effects were not consistently more pronounced in speech maskers than noise maskers.