News:
- NEW! Groundtruth for test set released. (29-3-23)
- Ranking tables now available (12-2-23)
- Instructions for preparing your paper now available (9-2-23)
- Important!: you can request a sample test format file (1-2-23)
- Deadline for submission of results extended to 9 February (1-2-23).
- ADReSS-M test metadata (age, gender and education) released. Click here to download. (1-2-23).
- ADReSS-M test set released. Click here to download. (24-1-23)
- Challenge description and Baseline paper now available! (16-1-23)
- Registration deadline extended to 13 January (6-1-23).
- Important!: Test set Change; we will now use a test set of Greek spontaneous speech picture descriptions. A sample set will be provided. If you have already received the Spanish sample set, please ignore it and request the replacement Greek sample (6-1-23)
- Name Change: this SPGC is now named ADReSS-M (6-1-23)
- The Multilingual Alzheimer's Dementia Recognition on Spontaneous Speech, an ICASSP'23 Signal Processing Grand Challenge (SPGC) is announced! (27-11-22)
The SPGC on Multilingual Alzheimer's Dementia Recognition through Spontaneous Speech, at ICASSP 2023 The ADReSS-M Signal Processing Grand Challenge targets a difficult automaticprediction problem of societal and medical relevance, namely, thedetection of Alzheimer's Dementia (AD). Dementia is a category ofneurodegenerative diseases that entails a long-term and usuallygradual decrease of cognitive functioning. While there has been much interest in automated methods forcognitive impairment detection by the signal processing and machinelearning communities (de La Fuente, Ritchie and Luz, 2020), most of the proposed approaches have not investigated which speechfeatures can be generalised and transferred across languages forAD prediction, and to the best of our knowledge no work hasinvestigated acoustic features of the speech signal in multilingual ADdetection. The ADReSS-M Challenge targets this issue bydefining a prediction task whereby participants train theirmodels based on
English speech data and assess their models'performance on spoken
Greek data. It is expected that the modelssubmitted to the challenge will focus on acoustic features of thespeech signal and discover features whose predictive power ispreserved across languages, but other approaches can be considered. In keeping with the objectives of AD prediction evaluation, theADReSS-M challenge's dataset is statistically matched so as to mitigatecommon biases often overlooked in evaluations of AD detection methods,including repeated occurrences of speech from the same participant(common in longitudinal datasets), variations in audio quality, andimbalances of gender, age and educational level. By focusing on ADrecognition using spontaneous speech, we depart fromneuropsychological and clinical evaluation approaches, as spontaneousspeech analysis has the potential to enable novel applications forspeech technology in longitudinal, unobtrusive monitoring of cognitivehealth.This challenge aims to provide a platform for contributions anddiscussions on applying signal processing and machine learning methodsfor Multilingual Alzheimer's Dementia Recognition through SpontaneousSpeech. We invite the submission of papers describing system ideas(machine learning architectures), novel signal processing features,feature selection, and feature extraction methods in the context ofADReSS-M Challenge. The organizers have recently conducted an extensive systematic reviewof the scientific literature on speech and language processing AImethods for detection of cognitive decline in AD(deLa Fuente, Ritchie and Luz, 2020) which might offerparticipants a handy overview of approaches and results in this field.
