India, Oct. 30 -- Voice AI can speak many languages, but sounding natural in a real customer conversation is a different challenge. Regional accents, mixed-language sentences and local names can trip up generic speech models, especially in markets such as India.

Bengaluru-based Arrowhead AI is taking a more local approach with Arrowhead Fletch, its in-house text-to-speech (TTS) model built for multilingual customer conversations across India and Southeast Asia. The model supports 14 languages and, according to the company, already powers more than one million calls a day.

Built for the way people actually speak Fletch is designed to handle conversations that move between languages mid-sentence. This matters in India, where a customer in...