Introduction
We collaborated on a project aimed at developing an Italian-language speech recognition system, with a particular focus on regional accent diversity. The goal was to collect audio recordings of spontaneous conversations, varied by topic, speaker gender, and geographic origin. The challenge was to ensure both variety and authenticity while maintaining high quality standards and adhering to the technical specifications provided by the client.
Solution and Benefits
- Organization of collaborator pairs tasked with recording spontaneous conversations
- 140 minutes of recording per pair, split into sessions of at least 35 minutes each
- Predefined topics provided by the client to ensure semantic consistency and variety
- Balanced data collection in terms of gender and geographic origin, fully aligned with quality standards
Results and Conclusions
- Project completed in approximately 3 months
- Optimal balance between participant gender and topic diversity
- Full compliance with the client’s requirements for duration, spontaneity, and quality
- Delivery of a highly diverse dataset, ready for training advanced voice systems
