Alzheimer’s disease (AD) affects speech and language (Cummings, Benson, Hill & Read, 1985). This can be caused by the disease affecting language as a neurocognitive function itself or through other impaired neurocognitive functions that are also reflected in speech. AD speech is characterized at various levels, such as syntactic, semantic, fluency and lexical, as well as lexical retrieval difficulties (Slegers, Filiou, Montembeault & Brambati, 2018; Kavé & Goral, 2017). Irregularities in speech and language are useful predictors of disease progression (Forbes, Venneri & Shanks, 2002), giving rise to the idea that changes in speech features could be suitable for diagnostics. The study of speech biomarkers allows a valuable improvement in diagnostics beyond the clinical gold standard.
There are multiple reasons why speech biomarkers could improve clinical assessments.
First, speech biomarkers can capture the transition from pre-clinical stages to AD earlier than manually evaluated classical neuropsychological tests (König et al., 2015; Linz et al., 2019). Using linguistic (semantic, syntactic, morphological) and paralinguistic features (articulatory, phonetic), a successful differentiation between healthy control subjects and MCI patients can already be made (König et al., 2015). For this, speech from established and valid test procedures with verbal response format, such as verbal learning tasks, verbal fluencies or picture descriptions are usually used. All of these tests are sensitive measurement tools regarding Alzheimer’s (Pakhomov, Eberly & Knopman, 2018; Mueller, Hermann, Mecollari & Turkstra, 2018; Russo et al., 2016). However, studies have shown that early detection is even more accurate when speech features are additionally assessed. And the early differential diagnostic could even be detected before the AD stage: For example, applying a temporal analysis on the semantic verbal fluency task allows for a distinction between healthy persons and MCI patients (Linz et al., 2019).
Secondly, an examination with vocal biomarkers can be performed remotely, making the examination easier for the patient and reducing the burden at the clinical sites. Moreover, a higher sampling rate of assessments within a given period of time can be achieved, which allows a more fine-grained and within-subject personalized monitoring of disease progression. A low-tech solution is the examination by telephone, which can be based on the semantic verbal fluency task. The phone-based evaluation of word count, clusters and switches, and semantic metrics results in good classification quality (Tröger, Linz, König, Robert & Alexandersson, 2018).
Thirdly, the aforementioned experiment by Tröger et al. (2018) demonstrates that speech biomarkers provide the opportunity for automated analysis. This reduces the time required for neurocognitive testing and thus relieves the burden on the health care system. Their result shows that high accuracy of 89% for the classification of healthy and AD patients has been obtained. In addition, automatically extracted temporal and semantic clusters, and category switches from the semantic verbal fluency task were highly correlated with manually extracted values and led to accurate classification of healthy controls, MCI and AD patients (König et al., 2018). These findings demonstrate the feasibility of automated speech analysis in the detection of AD.
In conclusion, speech biomarkers can help to improve diagnostics in Alzheimer’s disease. They allow for an early diagnosis of the disease and can be used remotely and automatically. Several tasks with verbal response format are suitable for speech analysis. Especially the semantic verbal fluency task, which is easy to administer, is widely supported. Future work aims to further validate speech biomarkers for use in early detection during the pre-clinical stages of dementia.