As part of a much larger language coverage, GeoPoll provides pre-labeled, high-quality Portuguese-language audio datasets purpose-built for training artificial intelligence models. Unlike synthetic or scripted datasets, our data is sourced from real phone interviews conducted with native speakers across multiple countries. These interviews are structured using domain-specific scripts for thematic consistency while allowing for spontaneous, natural responses.

Each recording is transcribed and diarized by human linguists fluent in local Portuguese variants, then tagged with rich metadata including age, gender, dialect, and location. The result is a scalable library of real-world Portuguese conversations, optimized for use in LLM fine-tuning, ASR training, TTS synthesis, and multilingual AI applications.