Embedding Culturally Responsive Voice Recognition in Community Health Ecosystems: A Participatory AI Approach
No Thumbnail Available
Authors
Dr Pascal Landindome Navelle
Dr Shaun Danquah
Contact
Check for full-text access
Issue Date
06-May-26
Type
Conference Abstract
Language
Keywords
Working with people and communities , Economic inclusion , Implementation/scale up , Embedded researchers , Neighbourhood health & place-based working
Alternative Title
Abstract
Background
Voice recognition technologies are increasingly integrated into NHS and local authority–supported digital pathways including digital triage, ambient voice technology (AVT), and remote monitoring. However, evidence demonstrates substantial disparities in speech recognition accuracy across accents, dialects, and racialised speech patterns, contributing to risks of widened health inequalities [1–3]. These disparities are especially concerning in safety critical contexts such as telephone triage, where clinical decisions rely heavily on accurate verbal communication [1]. Studies also show marked differences in transcription accuracy between Black and White patients in clinical interactions [4].
Lambeth Council, through the Health Determinants Research Collaboration (HDRC), identified these inequities as a priority for local authority leadership in the ethical governance, evaluation, and safe adoption of AI tools.
Objectives
This project aims to develop a community embedded, participatory framework for designing, evaluating, and governing culturally responsive voice AI systems. The objectives are to:
1. Examine ASR performance disparities across Lambeth's culturally and linguistically diverse population.
2. Co produce representative, community sourced speech datasets with young people and seldom heard communities.
3. Develop an ethical and governance model aligned with Lambeth Council safeguarding duties and NHS AVT guidance [5].
Stage at Submission
Early development. Lambeth HDRC and community partners have completed:
• scoping of priority digital service areas (e.g., access lines, navigation tools),
• partnership building, and
• identification of community groups needed to shape equitable dataset design and governance.
Methods
A participatory, community led design methodology including:
• Co production workshops with young people and minoritised communities;
• Technical audit of existing ASR systems using locally sourced speech samples;
• Community driven data curation aligned with equity centred AI standards [2,3];
• Governance mapping incorporating Lambeth's safeguarding processes and NHS AI/AVT standards [5].
Learning So Far / Discussion Points
Early engagement shows strong demand for local authority leadership in AI ethics and clear community concerns around digital exclusion. There is emerging potential for cross borough collaboration on dataset development, governance standards, and evaluation models across London. The presentation will explore opportunities for shared learning, scaling, and aligning procurement decisions with community rooted ethical frameworks.