Company
Date Published
Author
Russell Kahan
Word count
2761
Language
English
Hacker News points
None

Summary

Twilio provides several best practices to improve speech recognition performance in their IVR and Virtual Agent applications, including utilizing mobile devices for better audio quality, choosing a high-quality PSTN connection provider, leveraging "Hints" in the `<Gather>` verb, using Enhanced Speech Recognition with Twilio's recommended Google ASR speech model, engineering prompts to encourage natural and clear speech input, keeping it clean by filtering profanities, offering Dual-Tone Multi-frequency (DTMF) as a backup input method when speech recognition fails, streaming speech data to an external speech recognition provider, leveraging confidence scoring in the prompting application, and implementing 2FA or other post-processing techniques to deal with near-homonyms. By following these best practices, developers can maximize the odds of a superior automated self-service experience with Twilio.