Aphasia is a language disorder that affects millions of adults worldwide annually; it is most commonly caused by strokes or neurodegenerative diseases. Anomia, or word finding difficulty, is a prominent symptom of aphasia, which is often diagnosed through confrontation naming tasks. In the clinical setting, identification of correctness in responses to these naming tasks is useful for diagnosis, but currently is a labor-intensive process. This year’s Post-Stroke Speech Transcription Challenge provides an opportunity to explore ways of automating this process. In this work, we focus on Task B of the challenge, i.e. identification of response correctness. We study whether a simple aggregation of using the 1-best automatic speech recognition (ASR) output and acoustic features could help predict response correctness. This was motivated by the hypothesis that acoustic features could provide complementary information to the (imperfect) ASR transcripts. We trained several classifiers using various sets of acoustic features standard in speech processing literature in an attempt to improve over the 1-best ASR baseline. Results indicated that our approach to using the acoustic features did not beat the simple baseline, at least on this challenge dataset. This suggests that ASR robustness still plays a significant role in the correctness detection task, which has yet to benefit from acoustic features.

Analysis of client and therapist behavior in counseling sessions can provide helpful insights for assessing the quality of the session and consequently, the client’s behavioral outcome. In this paper, we study the automatic classification of standardized behavior codes (annotations) used for assessment of psychotherapy sessions in Motivational Interviewing (MI). We develop models and examine the classification of client behaviors throughout MI sessions, comparing the performance by models trained on large pretrained embeddings (RoBERTa) versus interpretable and expert-selected features (LIWC). Our best performing model using the pretrained RoBERTa embeddings beats the baseline model, achieving an F1 score of 0.66 in the subject-independent 3-class classification. Through statistical analysis on the classification results, we identify prominent LIWC features that may not have been captured by the model using pretrained embeddings. Although classification using LIWC features underperforms RoBERTa, our findings motivate the future direction of incorporating auxiliary tasks in the classification of MI codes.

In conversational speech, the acoustic signal provides cues that help listeners disambiguate difficult parses. For automatically parsing spoken utterances, we introduce a model that integrates transcribed text and acoustic-prosodic features using a convolutional neural network over energy and pitch trajectories coupled with an attention-based recurrent neural network that accepts text and prosodic features. We find that different types of acoustic-prosodic features are individually helpful, and together give statistically significant improvements in parse and disfluency detection F1 scores over a strong text-only baseline. For this study with known sentence boundaries, error analyses show that the main benefit of acoustic-prosodic features is in sentences with disfluencies, attachment decisions are most improved, and transcription errors obscure gains from prosody.