cs.AIMar 24, 2026

Evaluating a Multi-Agent Voice-Enabled Smart Speaker for Care Homes: A Safety-Focused Framework

Authors: Zeinab Dehghani, Rameez Raja Kureshi, Koorosh Aslansefat, Faezeh Alsadat Abedi, Dhavalkumar Thakker, Lisa Greaves, Bhupesh Kumar Mishra, Baseer Ahmad, 

Organizations: University of Hull, UK · University of Southampton, UK · Connexin, Hull, UK · Leeds Teaching Hospital NHS Trust, UK

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Artificial intelligence (AI) is increasingly being explored in health and social care to reduce administrative workload and allow staff to spend more time on patient care. This paper evaluates a voice-enabled Care Home Smart Speaker designed to support everyday activities in residential care homes, including spoken access to resident records, reminders, and scheduling tasks. A safety-focused evaluation framework is presented that examines the system end-to-end, combining Whisper-based speech recognition with retrieval-augmented generation (RAG) approaches (hybrid, sparse, and dense). Using supervised care-home trials and controlled testing, we evaluated 330 spoken transcripts across 11 care categories, including 184 reminder-containing interactions. These evaluations focus on (i) correct identification of residents and care categories, (ii) reminder recognition and extraction, and (iii) end-to-end scheduling correctness under uncertainty (including safe deferral/clarification). Given the safety-critical nature of care homes, particular attention is also paid to reliability in noisy environments and across diverse accents, supported by confidence scoring, clarification prompts, and human-in-the-loop oversight. In the best-performing configuration (GPT-5.2), resident ID and care category matching reached 100% (95% CI: 98.86-100), while reminder recognition reached 89.09% (95% CI: 83.81-92.80) with zero missed reminders (100% recall) but some false positives. End-to-end scheduling via calendar integration achieved 84.65% exact reminder-count agreement (95% CI: 78.00-89.56), indicating remaining edge cases in converting informal spoken instructions into actionable events. The findings suggest that voice-enabled systems, when carefully evaluated and appropriately safeguarded, can support accurate documentation, effective task management, and trustworthy use of AI in care home settings.

Figures & tables

Figure 1: System overview and architecture of the voice-enabled care support platform.
Figure 2: Assurance case for the Care Home Smart Speaker. The metrics-based argument A2 justifies the parsing, inserting, and scheduling risk C1.5 using three KPIs evaluated with 95% Wilson confidence intervals: C2.3 database insertion accuracy, C2.4 reminder scheduling success, and C2.5 absence of hallucinations during parsing. Retrieval risk ( C1.1 ) is decomposed by A1 into standard and paraphrased queries, supported by evidence E2.1–E2.3 and defeater D2.2 .
Evaluation ComponentConfiguration / ScenarioData SourceNo. of InstancesLogged Outputs
Care-home trial contextSupervised pilot trials conducted in two UK residential care homes (NLG and Beverley Grange, Hull). Interactions occurred in typical staff areas with background noise and interruptions. The system operated strictly as a decision-support tool.Care-home staff interactions2 sitesTrial logs, qualitative observations
Participants and interaction scenariosCare staff interacted with the system using natural speech. Scenarios included care documentation, note recording, care information queries, and reminder scheduling. Both scripted and unscripted interactions were included.Spoken interactions330 transcriptsAudio recordings, transcripts, parsed outputs
Speech recognition and preprocessingWhisper-based ASR with signal-quality safeguards and rejection of low-confidence audio segments to prevent unreliable transcriptions.Spoken input330 audio inputsAccepted/rejected audio segments, ASR transcripts
Parsing and database insertionTranscripts parsed into structured records containing resident ID, care category, timestamp, and free-text notes. Schema validation enforced before insertion into a PostgreSQL database.ASR transcripts330 parsed recordsStructured database entries, validation logs
Reminder extractionReminder descriptions extracted from spoken notes when reminder intent was detected.Reminder-containing transcripts184 casesExtracted reminder text, classification outputs
Retrieval evaluationSpoken queries issued by care staff and evaluators using sparse, dense, and hybrid retrieval strategies. Needle-in-a-haystack tests embedded critical records within larger document sets.Trial + synthetic database5 representative queriesRetrieved records, ranking positions, similarity scores
Table 1: Summary of the experimental setup used to evaluate the Care Home Smart Speaker system.
MetricModelAccuracy (%)CI Lower (%)CI Upper (%)
Category MatchGPT 5.210098.86100
 LLaMA–397.8795.7799.08
 Qwen67.8862.6672.69
Resident ID MatchGPT 5.210098.86100
 LLaMA–397.8795.6898.96
 Qwen66.3661.1071.25
Table 2: Database insertion and reminder recognition accuracy across models with 95% Wilson confidence intervals.
ModelTPFPFNTN
GPT 5.2184360110
LLaMA–3131053146
Qwen107077146
Table 3: Confusion matrix (TP/FP/FN/TN) for reminder recognition across models.
FieldMetricModelnMeanP25P75
ReminderCosine DistanceGPT 5.21840.07430.02740.1022
ReminderCosine DistanceLLaMA–31270.10310.04060.1410
ReminderCosine DistanceQwen1060.16300.09890.2013
ReminderWMDGPT 5.21840.32420.20590.4460
ReminderWMDLLaMA–31270.42360.24620.5690
ReminderWMDQwen1060.54760.42660.6557
Table 4: Distance summary between ground-truth and system-generated text. Lower values indicate higher semantic similarity. P25 and P75 denote the 25th and 75th percentiles.
Figure 3: Per-category accuracy for GPT 5.2, showing category matching, resident ID matching, and reminder recognition across care categories.
Figure 4: Per-category accuracy for LLaMA–3, showing category matching, resident ID matching, and reminder recognition across care categories.
Figure 5: Per-category accuracy for Qwen, showing category matching, resident ID matching, and reminder recognition across care categories.
MetricModelAccuracy (%)CI Lower (%)CI Upper (%)
Category MatchGPT 5.210097.95100.00
 LLaMA–398.1896.0999.16
 Qwen67.8860.7974.14
Resident ID MatchGPT 5.210097.95100.00
 LLaMA–398.1896.0999.16
 Qwen66.3659.2572.79
Table 5: Google Calendar integration accuracy with 95% Wilson confidence intervals across models.
MethodMetricModelMean
Hybrid RAGCosine DistanceGPT 5.20.371541
  LLaMA–30.428666
  Qwen0.418436
Hybrid RAGWMD DistanceGPT 5.20.743192
  LLaMA–30.883987
  Qwen0.950280
Table 6: Mean retrieval distances across RAG methods per model. Lower distances indicate closer alignment with ground truth.
More in this topic

Voice Agent Evaluation

Voice agent evaluation measures how reliably spoken AI systems understand, respond, take actions, and recover in real interactions.

See all 45 papers

Explore similar work

CardsList
May 8, 2026

AI-Care: A Conversational Agentic System for Task Coordination in Alzheimer's Disease Care

Individuals with Alzheimer's disease (AD) and Alzheimer's disease-related dementia (ADRD) experience memory and thinking changes that impact their ability to use digital daily management tools. For example, adding an event to a digital calendar requires multiple steps that may act as barriers to independent use for individuals with AD/ADRD. This paper presents AI-Care, a conversational agentic artificial intelligence (AI) layer built on top of a remote caregiving platform co-designed with people with AD/ADRD. AI-Care is designed to reduce the cognitive load on individuals with AD/ADRD when managing everyday tasks such as setting calendar reminders and organizing to-do lists through natural-language interaction with a voice-first chatbot. The system uses a LangGraph-based stateful orchestration approach in which each request passes through sanitization, intent classification, context loading, safety checks, deterministic slot collection, tool execution, and response composition. Safety-critical responses, particularly around medications and allergies, are grounded in caregiver-verified records rather than free-form model generation. The system does not make autonomous medical or treatment decisions. Incomplete or ambiguous requests are handled through controlled multi-turn clarification rather than silent failure or guessing. The system supports both typed and spoken input, with voice output through ElevenLabs text-to-speech. Longer responses are chunked before synthesis to avoid rushed playback. A preliminary pilot with four individuals with mild-to-moderate AD/ADRD showed that users found the system trustworthy, competent, and likable, and were able to complete the evaluated coordination tasks through conversation. We describe the design goals, system architecture, safety controls, and findings from this formative evaluation.

Preyash Yadav, Michelle Cohn, Priyanka Koppolu +5

Department of Computer Science University of California, Davis Davis, USA · Department of Linguistics University of California, Davis Davis, USA · Department of Neurology University of California, Davis Sacramento, USA +1

Alzheimer's DiseaseTool-Using AgentsConversational AgentsLLM Agent SafetyDementia Care
AskPDF
Aug 6, 2026

ECHO: A Locally-Deployable Agentic Health Assistant with Temporal Memory, Safety Guardrails, and Speech Assessment

This paper presents ECHO (Enhanced Care & Health Observer), a locally-deployable conversational health assistant for long-term chronic care management. ECHO integrates three complementary software modules developed under shared supervision as a unified system. The core module is an agentic chatbot built on a ReAct loop orchestrated via LangGraph, equipped with 17 clinical tools and a temporal knowledge graph for persistent cross-session memory; it achieves a 94.9% tool-execution pass rate across a 59-scenario benchmark with GPT-5 Mini. A two-stage hybrid safety layer intercepts all incoming queries: a rule-based layer handles explicit crisis signals and jailbreak attempts in under 1ms, while a signed graph neural network (GNN) with APPNP-style propagation classifies boundary cases by clinical intent, achieving 88.8% accuracy and 90.6% unsafe recall on a 2,537-query annotated Turkish health dataset while outperforming zero-shot LLM baselines including Llama 3.3 70B. A multimodal speech assessment module combining Whisper acoustic encoding and BERT text encoding with cross-attention fusion estimates emotion, depression, and pain, reaching a mean macro F1 of 0.652. The full system is implemented as a web application that can run entirely on consumer hardware, with no patient data transmitted to external services, supporting compliance with GDPR and KVKK.

Abdulkadir Külçe, Alihan Esen, Çağla Fikir +4

Dept. of Artificial Intelligence and Data Engineering, Istanbul Technical University, Istanbul, Turkey · Dept. of Computer Engineering, Istanbul Technical University, Istanbul, Turkey

Conversational AgentsLLM Agent SafetyMemory-Augmented Language Model Agents
AskPDF
Mar 26, 2026

Back to Basics: Revisiting ASR in the Age of Voice Agents

Automatic speech recognition (ASR) systems have achieved near-human accuracy on curated benchmarks, yet still fail in real-world voice agents under conditions that current evaluations do not systematically cover. Without diagnostic tools that isolate specific failure factors, practitioners cannot anticipate which conditions, in which languages, will cause what degree of degradation. We introduce WildASR, a multilingual (four-language) diagnostic benchmark sourced entirely from real human speech that factorizes ASR robustness along three axes: environmental degradation, demographic shift, and linguistic diversity. Evaluating seven widely used ASR systems, we find severe and uneven performance degradation, and model robustness does not transfer across languages or conditions. Critically, models often hallucinate plausible but unspoken content under partial or degraded inputs, creating concrete safety risks for downstream agent behavior. Our results demonstrate that targeted, factor-isolated evaluation is essential for understanding and improving ASR reliability in production systems. Besides the benchmark itself, we also present three analytical tools that practitioners can use to guide deployment decisions.

Geeyang Tay, Wentao Ma, Jaewon Lee +8

Boson AI

Robust ASRAutomatic Speech RecognitionASR EvaluationMultilingual ASRDemographic Bias in ASR
AskPDF