Building a Robust Voicemail Detection System at Bland

Voicemail detection in telephony presents significant challenges due to the lack of standardization across carriers and devices, complicating AI's ability to accurately discern when a call has reached voicemail. Unlike humans who can intuitively identify voicemail, AI systems risk misclassification, leading to unintended messages and inefficiencies in call centers. To address this, Bland developed advanced machine learning models for voicemail detection, utilizing fine-tuned Wave2Vec and CNN models that achieved high accuracy rates of 98.5% and 97%, respectively. These models help streamline AI call center operations by filtering out calls that hit voicemail, thus enhancing engagement and conversion rates. Bland is also exploring a novel approach using silence detection to further improve voicemail detection accuracy, aiming to distinguish between automated systems and live interactions. Both models have been open-sourced for community exploration, though the training data remains closed to protect privacy, indicating an ongoing commitment to evolving these systems in line with changing voicemail technologies.

No tracked trend matches for this post yet.