Refine
Document Type
- Conference Paper (3)
- Article (2)
Institute
Has Fulltext
- yes (5)
Is part of the Bibliography
- yes (5)
Keywords
This paper introduces aphaDIGITAL – a mobile application for speech and language therapy (SLT) support of German-speaking people with aphasia. The app includes automatic speech recognition and text analysis components, which allows giving detailed feedback to the users on different types of errors. Furthermore, the exercises are accompanied by an avatar-based SLT assistant, which is created for this specific purpose. The user performs the exercises, individually configured for her by the SLT practitioner, on a mobile device. Data collection and processing are carried out on the server.
Those who suffer from aphasia benefit from digital speech and language therapy solutions, and automatic speech recognition (ASR) has been already used for giving feedback on the correctness of the answers in naming exercises. AphaDIGITAL application is to provide German-speaking users with detailed feedback on phonemic/phonetic and semantic errors, based on automatic speech and language processing. For this purpose, open-source ASR solutions for German were evaluated on different corpora of atypical speech, including two small datasets with aphasic speech samples. Character error rate, the number of precisely recognized items and empty outputs served as evaluation metrics. The four selected models are generally robust to the deteriorated condition of speech and audio quality and consistently outperform commercial models in atypical speech recognition. Applying error acceptance threshold, additional use of phonemic error rate, and other valuable insights for ASR implementation in aphaDIGITAL are discussed.
Post hoc implementation of non-standard phonetic features in the context of aphasic speech analysis
(2025)
Despite current progress, automatic speech recognition (ASR) often struggles with non-standard speech, for example, influenced by dialectal or pathological features. (Re)training ASR models to accommodate these variations is not always possible due to limited data. This paper proposes applying the knowledge about non-standard (aphasic and dialectal) phonetic features to the ASR transcription post hoc. Using speech data from German speakers with aphasia who speak the Thuringian-Upper Saxon dialect, this study evaluates the impact of these modifications on an ASR-based error analysis pipeline. The approach helps to reduce automatic error rates on the recordings manually labelled as error-free. The performance of the pipeline also improves both in general acceptance or rejection of the responses and error attribution. General acceptance/rejection accuracy reaches the mean of 83.3%, which is considered sufficient to be used in a digital application for speech and language therapy support.
Linguistic and extralinguistic factors in automatic speech recognition of German atypical speech
(2024)
Automatic speech recognition (ASR) has been already used in speech and language therapy, including diagnostic tasks and practice exercises for people with aphasia (PWA). The lack of relevant data makes it difficult to evaluate the algorithms’ suitability for German-speaking PWA. For the current project, four open-source ASR models were selected based on their performance on other types of atypical speech, and the details of their evaluation are presented in this paper. The four selected models are generally robust to speakers’ gender and age. The one-word recognition yields better results for words of moderate length. Speech rate should be neither too slow nor too quick for lower error rates both in words and phrases, and the latter should be also of moderate length.
Diese Studie analysiert, ob Gesichtsausdrücke in generierten Bildern genauso wahrgenommen werden wie in realen Bildern. Dabei werden die Wahrnehmung der Emotionsintensität und Sympathie, die Fähigkeit zur Emotionserkennung sowie die Unterscheidung zwischen echten und generierten Bildern untersucht. Zur Beantwortung dieser Fragen wurde eine Online-Umfrage mit mehr als 200 Teilnehmern durchgeführt. Die Ergebnisse zeigen geringfügige bis moderate Unterschiede in der Wahrnehmung zwischen generierten und realen Bildern, allerdings wird auch deutlich, dass die Unterscheidung zwischen beiden zunehmend schwieriger wird.
