Main content

Home

Menu

Loading wiki pages...

View
Wiki Version:
We developed a web-hosted system for speech recognition testing and training with spoken word and sentence materials that are visual-only (VO, lipreading), auditory-only (AO), and audiovisual (AV). Stimuli may be presented in quiet and in noise. Screening may be conducted for hearing thresholds, vision, and lipreading. The software was originally intended for web-based applications and was enhanced due to Covid-19. The system software can support flexible and efficient development of new protocols for training and testing. Applications are available through a browser interface. A central feature of the system is its unique software component for scoring open set recognition responses with sentence stimuli. In real time, the software computes words correct/incorrect, phonemes correct/incorrect, phonetic perception measures, and measures of inserted words or phonemes. The real-time capability has allowed us to develop unique approaches to feedback for speech recognition training. We will describe the overall system capabilities and highlight several applications. The lipreading training procedures we are now testing remotely were shown under laboratory conditions to promote significant improvements in VO, and AO and AV in noise speech recognition with generalization to untrained open set stimuli. The lipreading training procedures have been used as a teaching tool for graduates and undergraduates in audiology classes. The software scoring capabilities have also been shown to be effective for mining individual differences data from responses to an existing clinical test of speech recognition in noise. The data mining has shown that incorrect open set responses disclose information about a listener’s phonetic perceptual fidelity and their higher-level ability to exclude babble noise during open set responses to sentences. lbernste@gwu.edu May 3, 4 or 5