[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-469157-105":59,"doc-detail-469157-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","ai-powered-speech-analysis-automating-transcription-embeddings-and-deep-learning-for-early-alzheimers-detection","AI-Powered Speech Analysis - Automating Transcription, Embeddings, and Deep Learning for Early Alzheimer’s Detection","","Spontaneous speech serves as a promising non-invasive biomarker for early Alzheimer’s detection. The study transcribes participant audio, generates large language model (LLM) vector embeddings, and trains machine learning classifiers to distinguish Alzheimer’s disease (AD) patients from healthy controls. Audio comes from the ADReSS 2020 Challenge, describing the Cookie Theft picture. Linguistic features and filler-word/disfluency analyses are compared with embedding-based approaches, including ROC improvements and accuracy/precision results across SVM, logistic regression, and random forest.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":35,"@type":76,"position":81},"https://docshare.wps.com/document/healthcare/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/ai-powered-speech-analysis-automating-transcription-embeddings-and-deep-learning-for-early-alzheimers-detection/469157/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/ai-powered-speech-analysis-automating-transcription-embeddings-and-deep-learning-for-early-alzheimers-detection/469157.png","ImageObject",300,407,{"name":92,"@type":93},"Marglet","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-07","2026-09-30",true,{"@type":102,"interactionType":103,"userInteractionCount":14},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What data and task setup were used in the study?","Question",{"text":112,"@type":113},"The study used audio files from the ADReSS 2020 Challenge dataset. Participants described the Cookie Theft picture, with training and testing groups composed of AD patients and healthy controls.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How did the researchers obtain transcriptions and embeddings?",{"text":117,"@type":113},"Audio was transcribed using two commercial audio-to-text APIs, OpenAI Whisper and AssemblyAI. The approach then generated embedding vectors using two OpenAI embedding models and used these embeddings for classification.",{"name":119,"@type":110,"acceptedAnswer":120},"Which classification model performed best and what metrics were reported?",{"text":121,"@type":113},"The Support Vector Machine (SVM) model outperformed logistic regression and random forest. It reported about 0.84 accuracy and 0.83 precision.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},469157,1790775179,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":34,"category_name":35,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":14,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":14,"language":139,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":140,"faqs":141,"seo_title":142,"seo_description":67,"update_tm":129,"read_time":24,"preview_page_count":4},2336477982702,"https://ap-avatar.wpscdn.com/davatar_9964176cb1d06d4a9deccf72a44ae3dc","DOI: 10. 1002/alz70856_107467  \nBIOMARKERS  \nPOSTER PRESENTATION  \nBIOMARKERS (NON-NEUROIMAGING)  \nAI-Powered Speech Analysis: Automating Transcription, Embeddings, and Deep Learning for Early Alzheimer’s Detection  \nDaniel Zhou 1  Zhen Qian2  \n1Skyline High School, Ann Arbor, MI, USA  \n2 University of Michigan, Ann Arbor, MI, USA  \nCorrespondence  \nZhen Qian, University of Michigan, Ann Arbor, MI, USA.  \n[Email:](Email: zqian@umich.edu)[ zqian@umich.edu](Email: zqian@umich.edu)  \nAbstract  \nBackground: Spontaneous speech is a promising, non-invasive, cost-effective biomarker. LLM vector embeddings capture semantic and contextual patterns. This study transcribed audio, generated embeddings, and trained machine learning models to classify AD patients versus healthy controls.  \nMethod: We used audio files from the Alzheimer’s Dementia Recognition through Spontaneous Speech (ADReSS) 2020 Challenge dataset, with 108 training participants (54 AD, 54 healthy) and 48 testing participants, all describing the Cookie Theft picture. We did linguistic analysis for word frequency and speech disfluency. Next, we evaluated two commercial audio-to-text APIs (OpenAI Whisper vs AssemblyAI) to build a more automated and scalable classification pipeline. We used two OpenAI’s new embedding models to generate embedding vectors. Lastly we built three classification models: Support Vector Machine (SVC), Logistic Regression (LR) and Random Forest (RF) and compared model performance.  \nResult: We can find a higher frequency of filler words and fewer diversity of vocabulary in the word cloud of Alzheimer’s. The classification model’s ROC values are significantly improved when the LLM vector embedding is used, compared to the model based solely on linguistic features. Combining vector embeddings with linguistic features only resulted in marginal improvements. We used both the OpenAI Whisper and the AssemblyAI to generate audio transcriptions. We listed detailed metrics for the classification model trained with embedding vectors combined with linguistic features. The Support Vector Machine (SVM) model out-performed the Logistic Regression(LR) and Random Forest(RF) models with 0.84 inAccuracy, 0.83 in Precision.  \nConclusion: We have built an AI-based pipeline to transform spontaneous speech recordings into Large Language Model vector embeddings, and compared performance of different machine learning models for Alzheimer’s classification. The project source code is hosted on GitHub as an open source project: [https://github.com/dzhou08/](https://github.com/dzhou08/)  \nThis is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.  \n© 2025 The Alzheimer’s Association. Alzheimer’s & Dementia published by Wiley Periodicals LLC on behalf of Alzheimer’s Association.  \nAlzheimer’s Dement. 2025;21(Suppl. 2):e107467.  \n[https://doi.org/10.1002/alz70856_107467](https://doi.org/10.1002/alz70856_107467)  \nwi[leyonlinelibrary.com/journal/alz](leyonlinelibrary.com/journal/alz)  \n1of2  \nBIOMARKERS  \nembedding_AZ In the future, we will integrate this pipeline into a web application asa free service. We will also look into possibilities of combining speech embeddings with other multimodal biomarker data to further enhance the diagnostic accuracy. With the rapid advance ofAI and sensor technologies, multimodalframeworks will likely become standard in dementia care in the near future.","cbCaiskRSFP07b8v","https://ap.wps.com/l/cbCaiskRSFP07b8v","pdf",86407,"English","# BIOMARKERS\n## Background\n## Method\n## Result\n## Conclusion","[{\"question\":\"What data and task setup were used in the study?\",\"answer\":\"The study used audio files from the ADReSS 2020 Challenge dataset. Participants described the Cookie Theft picture, with training and testing groups composed of AD patients and healthy controls.\"},{\"question\":\"How did the researchers obtain transcriptions and embeddings?\",\"answer\":\"Audio was transcribed using two commercial audio-to-text APIs, OpenAI Whisper and AssemblyAI. The approach then generated embedding vectors using two OpenAI embedding models and used these embeddings for classification.\"},{\"question\":\"Which classification model performed best and what metrics were reported?\",\"answer\":\"The Support Vector Machine (SVM) model outperformed logistic regression and random forest. It reported about 0.84 accuracy and 0.83 precision.\"}]","AI-Powered Speech Analysis - Automating Transcription, Embeddings, and Deep Learning for Early Alzheimer’s Detection | PDF"]