Medical Device Promo Call Dataset – 300 Hours (US Customers & Filipino Agents)

This dataset comprises approximately 300 hours of real-world call center conversations focused on medical device promotions and customer engagement. The interactions occur between US-based customers and Filipino customer service or sales agents, reflecting authentic offshore call center operations. Each audio recording is paired with a structured JSON transcript, enabling seamless integration into machine learning pipelines for speech and natural language processing applications. The dataset captures natural conversational dynamics, including inquiries, product explanations, objections, clarifications, and closing discussions. It reflects realistic communication patterns such as interruptions, pauses, varying speech speeds, and accent diversity.

Data Quality

Accurate speech-to-text conversion with high alignment to audio Time-stamped segments enabling precise mapping between audio and text Cleaned transcripts with: Removal of irrelevant noise markers Standardized formatting and punctuation Conversational integrity preserved (pauses, fillers like “um”, “uh” where relevant)

Use Cases

Train and evaluate speech-to-text models on real call center audio Improve recognition for: US accents Filipino accents Mixed-accent conversations