Summary We are building a voice-first data collection platform designed to operate entirely over standard mobile phone calls, without smartphones, mobile data, app or internet access on the user side. The platform will be used to collect high-quality, culturally grounded speech datasets from contributors, including rural and low-connectivity communities that are typically excluded from AI data pipelines. This is not a simple IVR or call-recording system. It is a production-grade, telephony-native platform with structured workflows, quality control, contributor state management, and scalable backend architecture. The system will serve as foundational infrastructure for building speech, language, and agentic AI models for African languages. Summary -Pre-selected contributors dial a phone number using basic feature phones -They interact with the system entirely via voice prompts -The system plays recorded prompts that trigger spontaneous speech (e.g. in your local dialect, discuss how you do crop rotation during harmattan or what local drugs you use for a child's breathing ailment) -Instead of a speech prompt, the contributor may also receive guided sentences for re-voicing tasks (e.g. listen to this sentence and say it exactly in your local accent) -Each recording is automatically checked for quality via a parallel pipeline that checks SNR -Accepted recordings are stored with rich metadata -Rejected recordings trigger polite re-recording flows -Contributors can check their progress and status via voice -Everything works without apps, screens, or internet access Deliverables -A telephony integration layer capable of handling inbound calls, session management, concurrent callers, and audio capture -A multi-language voice welcome and navigation flow that works via speech (not UI) -A prompt management system that plays recorded audio prompts and allows replay when needed -Two data collection workflows: Prompt-driven spontaneous speech and Guided sentence re-voicing (4–5 second sentences, replayable) -Real-time audio quality checks (for example signal-to-noise ratio and basic acoustic validation) -Logic to accept, reject, or request re-recording of submissions -Structured storage of audio files with linkage to prompts and metadata -Contributor-level tracking of completed, approved, rejected, and pending recordings -Voice-based status reporting so contributors can hear their progress and standing -Safeguards against abuse, duplication, or low-effort submissions -An admin or internal interface to monitor collection progress, quality metrics, and system health