Concept: Speech recognition
The technology for evaluating patient-provider interactions in psychotherapy-observational coding-has not changed in 70 years. It is labor-intensive, error prone, and expensive, limiting its use in evaluating psychotherapy in the real world. Engineering solutions from speech and language processing provide new methods for the automatic evaluation of provider ratings from session recordings. The primary data are 200 Motivational Interviewing (MI) sessions from a study on MI training methods with observer ratings of counselor empathy. Automatic Speech Recognition (ASR) was used to transcribe sessions, and the resulting words were used in a text-based predictive model of empathy. Two supporting datasets trained the speech processing tasks including ASR (1200 transcripts from heterogeneous psychotherapy sessions and 153 transcripts and session recordings from 5 MI clinical trials). The accuracy of computationally-derived empathy ratings were evaluated against human ratings for each provider. Computationally-derived empathy scores and classifications (high vs. low) were highly accurate against human-based codes and classifications, with a correlation of 0.65 and F-score (a weighted average of sensitivity and specificity) of 0.86, respectively. Empathy prediction using human transcription as input (as opposed to ASR) resulted in a slight increase in prediction accuracies, suggesting that the fully automatic system with ASR is relatively robust. Using speech and language processing methods, it is possible to generate accurate predictions of provider performance in psychotherapy from audio recordings alone. This technology can support large-scale evaluation of psychotherapy for dissemination and process studies.
Extensive research shows that inter-talker variability (i.e., changing the talker) affects recognition memory for speech signals. However, relatively little is known about the consequences of intra-talker variability (i.e. changes in speaking style within a talker) on the encoding of speech signals in memory. It is well established that speakers can modulate the characteristics of their own speech and produce a listener-oriented, intelligibility-enhancing speaking style in response to communication demands (e.g., when speaking to listeners with hearing impairment or non-native speakers of the language). Here we conducted two experiments to examine the role of speaking style variation in spoken language processing. First, we examined the extent to which clear speech provided benefits in challenging listening environments (i.e. speech-in-noise). Second, we compared recognition memory for sentences produced in conversational and clear speaking styles. In both experiments, semantically normal and anomalous sentences were included to investigate the role of higher-level linguistic information in the processing of speaking style variability. The results show that acoustic-phonetic modifications implemented in listener-oriented speech lead to improved speech recognition in challenging listening conditions and, crucially, to a substantial enhancement in recognition memory for sentences.
Clinical documentation has undergone a change due to the usage of electronic health records. The core element is to capture clinical findings and document therapy electronically. Health care personnel spend a significant portion of their time on the computer. Alternatives to self-typing, such as speech recognition, are currently believed to increase documentation efficiency and quality, as well as satisfaction of health professionals while accomplishing clinical documentation, but few studies in this area have been published to date.
The HCAHPS Survey obtains hospital patients' experiences using four modes: Mail Only, Phone Only, Mixed (mail/phone follow-up), and Touch-Tone (push-button) Interactive Voice Response with option to transfer to live interviewer (TT-IVR/Phone). A new randomized experiment examines two less expensive modes: Web/Mail (mail invitation to participate by Web or request a mail survey) and Speech-Enabled IVR (SE-IVR/Phone; speaking to a voice recognition system; optional transfer to an interviewer). Web/Mail had a 12% response rate (vs. 32% for Mail Only and 33% for SE-IVR/Phone); Web/Mail respondents were more educated and less often Black than Mail Only respondents. SE-IVR/Phone respondents (who usually switched to an interviewer) were less often older than 75 years, more often English-preferring, and reported better care than Mail Only respondents. Concerns regarding inconsistencies across implementations, low adherence to primary modes, or low response rate may limit the applicability of the SE-IVR/Phone and Web/Mail modes in HCAHPS and similar standardized environments.
Feasibility of automated speech sample collection with stuttering children using interactive voice response (IVR) technology
- International journal of speech-language pathology
- Published about 3 years ago
Purpose: To investigate the feasibility of adopting automated interactive voice response (IVR) technology for remotely capturing standardized speech samples from stuttering children. Method: Participants were 10 6-year-old stuttering children. Their parents called a toll-free number from their homes and were prompted to elicit speech from their children using a standard protocol involving conversation, picture description and games. The automated IVR system was implemented using an off-the-shelf telephony software program and delivered by a standard desktop computer. The software infrastructure utilizes voice over internet protocol. Speech samples were automatically recorded during the calls. Video recordings were simultaneously acquired in the home at the time of the call to evaluate the fidelity of the telephone collected samples. Key outcome measures included syllables spoken, percentage of syllables stuttered and an overall rating of stuttering severity using a 10-point scale. Results: Data revealed a high level of relative reliability in terms of intra-class correlation between the video and telephone acquired samples on all outcome measures during the conversation task. Findings were less consistent for speech samples during picture description and games. Conclusions: Results suggest that IVR technology can be used successfully to automate remote capture of child speech samples.
Several behavioural studies have shown that the interplay between voice and face information in audiovisual speech perception is not universal. Native English speakers (ESs) are influenced by visual mouth movement to a greater degree than native Japanese speakers (JSs) when listening to speech. However, the biological basis of these group differences is unknown. Here, we demonstrate the time-varying processes of group differences in terms of event-related brain potentials (ERP) and eye gaze for audiovisual and audio-only speech perception. On a behavioural level, while congruent mouth movement shortened the ESs' response time for speech perception, the opposite effect was observed in JSs. Eye-tracking data revealed a gaze bias to the mouth for the ESs but not the JSs, especially before the audio onset. Additionally, the ERP P2 amplitude indicated that ESs processed multisensory speech more efficiently than auditory-only speech; however, the JSs exhibited the opposite pattern. Taken together, the ESs' early visual attention to the mouth was likely to promote phonetic anticipation, which was not the case for the JSs. These results clearly indicate the impact of language and/or culture on multisensory speech processing, suggesting that linguistic/cultural experiences lead to the development of unique neural systems for audiovisual speech perception.
Automatic speech processing (ASP) has recently been applied to very large datasets of naturalistically collected, daylong recordings of child speech via an audio recorder worn by young children. The system developed by the LENA Research Foundation analyzes children’s speech for research and clinical purposes, with special focus on of identifying and tagging family speech dynamics and the at-home acoustic environment from the auditory perspective of the child. A primary issue for researchers, clinicians, and families using the Language ENvironment Analysis (LENA) system is to what degree the segment labels are valid. This classification study evaluates the performance of the computer ASP output against 23 trained human judges who made about 53,000 judgements of classification of segments tagged by the LENA ASP. Results indicate performance consistent with modern ASP such as those using HMM methods, with acoustic characteristics of fundamental frequency and segment duration most important for both human and machine classifications. Results are likely to be important for interpreting and improving ASP output.
ABSTRACT Research with adults has shown that spoken language processing is improved when listeners are familiar with talkers' voices, known as the familiar talker advantage. The current study explored whether this ability extends to school-age children, who are still acquiring language. Children were familiarized with the voices of three German-English bilingual talkers and were tested on the speech of six bilinguals, three of whom were familiar. Results revealed that children do show improved spoken language processing when they are familiar with the talkers, but this improvement was limited to highly familiar lexical items. This restriction of the familiar talker advantage is attributed to differences in the representation of highly familiar and less familiar lexical items. In addition, children did not exhibit accent-general learning; despite having been exposed to German-accented talkers during training, there was no improvement for novel German-accented talkers.
Humans show a remarkable ability to understand continuous speech even under adverse listening conditions. This ability critically relies on dynamically updated predictions of incoming sensory information, but exactly how top-down predictions improve speech processing is still unclear. Brain oscillations are a likely mechanism for these top-down predictions [1, 2]. Quasi-rhythmic components in speech are known to entrain low-frequency oscillations in auditory areas [3, 4], and this entrainment increases with intelligibility . We hypothesize that top-down signals from frontal brain areas causally modulate the phase of brain oscillations in auditory cortex. We use magnetoencephalography (MEG) to monitor brain oscillations in 22 participants during continuous speech perception. We characterize prominent spectral components of speech-brain coupling in auditory cortex and use causal connectivity analysis (transfer entropy) to identify the top-down signals driving this coupling more strongly during intelligible speech than during unintelligible speech. We report three main findings. First, frontal and motor cortices significantly modulate the phase of speech-coupled low-frequency oscillations in auditory cortex, and this effect depends on intelligibility of speech. Second, top-down signals are significantly stronger for left auditory cortex than for right auditory cortex. Third, speech-auditory cortex coupling is enhanced as a function of stronger top-down signals. Together, our results suggest that low-frequency brain oscillations play a role in implementing predictive top-down control during continuous speech perception and that top-down control is largely directed at left auditory cortex. This suggests a close relationship between (left-lateralized) speech production areas and the implementation of top-down control in continuous speech perception.
During speech perception, linguistic elements such as consonants and vowels are extracted from a complex acoustic speech signal. Superior temporal gyrus (STG) participates in high-order auditory processing of speech, but how it encodes phonetic information is poorly understood. We used high-density direct cortical surface recordings in humans while they listened to natural, continuous speech to reveal the STG representation of the entire English phonetic inventory. At single electrodes, we found response selectivity to distinct phonetic features. Encoding of acoustic properties was mediated by a distributed population response. Phonetic features could be directly related to tuning for spectrotemporal acoustic cues, some of which were encoded in a non-linear fashion or by integration of multiple cues. These findings demonstrate the acoustic-phonetic representation of speech in human STG.