Abstract
<title>Abstract</title> <p>Traditional teacher-centered educational models in low- and middle-income countries (LMICs) often isolate learning outcomes from the underlying socio-emotional and resource-based processes that shape them. This study addresses these systemic measurement, experience, and voice gaps by exploring how artificial intelligence (AI)-driven, voice-mode student feedback can elevate student voice and redefine instructional practice. Guided by Self-Regulated Learning (SRL) theory, the Interactive, Constructive, Active, Passive (ICAP) cognitive framework, and Human-Centered Learning Analytics (HCLA), we conducted a mixed-methods investigation across seven secondary schools in Ibadan, Oyo State, Nigeria (N = 750). Quantitative survey data revealed critical barriers to mathematical achievement: 57% of students reported debilitating mathematics anxiety, 45% found current topics conceptually inaccessible, and 50% lacked essential home study materials. In parallel, a qualitative speech-based survey analyzed via a Natural Language Processing (NLP) sentiment analysis pipeline demonstrated that voice-mode responses bypass English-literacy barriers, mitigate student anxiety, and yield richer diagnostic insights than rigid, text-based Likert formats. By integrating students' spoken narratives into instructional decision-making, we demonstrate how real-time, AI-transcribed sentiment indicators shift teacher roles from "information deliverers" to "transformation partners" and elevate students from passive listeners to interactive co-creators of their learning environments. We discuss pedagogical, institutional, and policy implications for scaling human-centered AI feedback infrastructures to foster equitable and adaptive secondary education in Sub-Saharan Africa.</p>