Abstract
The identification of very short phoneme segments of speech is the key to understanding speech in chopped noise. Over the last 12 years, UIUC has repeated Miller-Nicely's 1955 phone recognition experiment, with 60 subjects and six SNRs, from quiet to −22 dB SNR, providing an improve understanding of the fundamentals of phone perception in normal (NH) and impaired (HI) listeners. Our phone robustness metric (SNR50) is the SNR such that the phone error is 50% (Toscano and Allen (2014), JSHLR). The error rate at SNR50 + 5 [dB] is <0.33%. We interpret this to mean that above SNR50, phonemes are below the Shannon channel-capacity limit. This is a game-changer: We must reevaluate speech recognition methods. For example, in an experiment on HI ears, we found that HI ears make large errors (e.g., 100%) on a small subset of tokens (Trevino Allen, JASA 134, 607, 2012). Averaging across tokens or listeners for any given consonant conflates the scores. There is a good news: Since there are only a small number of subject-dependent difficult sounds, testing time is reduced and accuracy is increased for a fixed test duration.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
