Dear Vapi Support Team, I hope you're doing well. We’ve been building an AI assistant using Vapi for our enterprise clients, and we’re currently facing a critical issue related to spoken address verification, which is essential to our workflow and tied closely to client privacy compliance. The challenge stems from the assistant’s difficulty in accurately recognizing spoken addresses, particularly due to variations in accents and pronunciations. Additionally, due to privacy constraints, the assistant is not permitted to reveal or read back the address on file — it can only prompt the user to repeat their address for confirmation. We’ve experimented with the custom keywords configuration, which allows us to set a probability threshold and essentially simulate fuzzy matching. However, this approach appears to only support single-word inputs, which limits its usefulness for real-world addresses. For example, a suburb like “South Wentworthville” would not be handled effectively using the current setup. Given these limitations, we’re looking for your guidance on: Whether there is a way to enable multi-word fuzzy matching, or If there’s an alternative method or best practice within Vapi for handling accurate, privacy-compliant address verification. We’re open to recommendations, provided the solution meets our privacy requirement of not exposing the stored address and can reliably handle multi-word entries in natural speech. Looking forward to your support and suggestions. Best regards,

Vapi doesn’t support multi-word fuzzy matching natively, but you can handle spoken address verification by boosting individual address components, using custom keywords, then performing fuzzy matching client-side to compare user input against the stored address without revealing it; optionally enhancing accuracy via an external address verification API for normalization and compliance.. https://docs.vapi.ai/customization/custom-keywords

thanks alot man this really helped