Skip to main content
. 2024 Mar 13;14:1260523. doi: 10.3389/fneur.2023.1260523

Table 2.

Examples of automated speech analysis in Alzheimer’s disease.

Publication Patient groups Main finding Methods
Orimaye et al. (2017) (108) Probable AD (n = 99)

  • Probable AD group had less use of syntactical components and greater use of lexical components in language compared to Healthy Controls (HCs).

  • Less use of n-grams (combinations or sequences of words that create a unit of meaning) in probable AD group than in HCs.

  • DementiaBank language transcript clinical dataset (111).

  • Automatic extraction of lexical, syntactic, and n-gram features of transcripts.
Healthy Controls (n = 99)
Yeung et al. (2021) (114) Healthy controls (n = 10)

  • Greater severity in word-finding difficulty and incoherence in MCI and AD compared to controls.

  • Automatically extracted features such as decreased word length and speech rate and increased pause frequency and length most strongly correlated with clinician ratings of WFD.

  • DementiaBank speech samples (111).

  • 5 clinicians blindly rated each speech sample on word finding difficulty, incoherence, perseveration, and speech errors, on a Likert scale from zero (nL) to 3 (severe impairment).

  • Automatic extraction of lexical, syntactic, semantic, and acoustic properties.
MCI (n = 10)
AD (n = 10)
Fraser et al. (JAD, 2016) (115) Healthy controls (n = 97)

  • Built a model which discriminates between HCs and possible/probable AD with 81% accuracy.

  • Semantic impairment, acoustic abnormality, syntactic impairment, and information impairment predict dementia diagnosis.

  • DementiaBank speech samples (111).

  • Considered 370 features including syntactic complexity, grammar, vocabulary richness, lexical content, repetitiveness, and acoustic.
Possible and Probable AD (n = 167)
Beltrami (Front. Aging Neurosci 2018) (116) Cognitively Impaired (n = 48: 32 MCI, 16 early dementia)

  • Acoustic features most altered in the patients compared to controls (including speech rate and pauses, and spectral properties).

  • Lexical features differentiate early dementia patients (e.g., fewer content words and modifiers).

  • Syntactic features (e.g., sentence complexity, fewer embedded phrases) decreased in early dementia and MCI patients.

  • Prospective study of spontaneous speech during description of a picture, typical working day, and last remembered dream.

  • Automatic extraction of lexical, rhythmic, acoustic, and syntactic features of speech.
Healthy Controls (n = 48)