Bengaluru-based voice AI company Arrowhead AI has launched Fletch, a text-to-speech (TTS) model it has built in-house for multilingual customer conversations across India and Southeast Asia. The model already sits inside the company’s live calling stack, where Arrowhead says it powers more than one million calls a day for lenders, banks, insurers and consumer brands.

The launch adds to a busy month for Indian speech AI, as homegrown companies race to own more of the voice layer rather than rent it from global APIs.

Built for code-mixed, number-heavy calls

Fletch supports 14 languages: English, 12 Indian languages including Hindi, Tamil, Telugu and Bengali, plus Malay and Indonesian. It can switch languages in the middle of a sentence, a common pattern in Indian customer calls where a payment reminder might mix Hindi and English while reading out an EMI amount and due date.

Arrowhead says the model starts producing audio in under 100 milliseconds and runs at about 80 milliseconds of latency, which matters for voice agents that need to sound conversational rather than robotic. It also offers 12 emotion styles, such as calm, empathetic, apologetic, firm and warm, so an agent can adjust its tone to what is being said.

The company reports roughly 30% better pronunciation accuracy on selected terms, including brand names, product names, customer names and Indian proper nouns. Arrowhead is careful to note that this figure applies to those terms specifically and is not a measure of overall call resolution or customer satisfaction.

Trained on regional speech and real deployments

Fletch is built on a DiTAR-based architecture (Diffusion Transformer Autoregressive Modeling for Speech Generation), which pairs language modelling with a diffusion transformer to generate continuous speech. Arrowhead adapted and trained it on thousands of hours of curated, use-case-specific regional speech, along with recordings from professional voice artists.

The model works alongside Arrowhead’s own speech recognition system, language model and a dedicated model for handling addresses. It is deployed across collections, sales, renewals and inbound customer service, and the company counts Axis Bank, RBL Bank, IndusInd Bank, Tata 1mg, upGrad, KreditBee, Kissht and TVS Credit among its clients.

“Owning our voice layer means we can tune every millisecond and every emotion for how India and Southeast Asia actually speak,” said Devyani Gupta, Founder and CEO of Arrowhead AI. “We want every customer to hear a conversation, not a machine.”

Co-founder and CTO Vengadanathan Srinivasan said the team built on recent research in speech generation and trained the model on regional speech with real voice artists, so it can adjust its tone in real time.

Why it matters

Voice is fast becoming the main interface for AI agents in Indian BFSI, where collections, renewals and sales still run largely on phone calls in local languages. Generic global speech models often stumble on regional accents, code-mixed sentences and Indian names, which forces customers to repeat themselves and erodes trust in automated calls.

By building its own TTS model, Arrowhead joins a growing group of Indian AI companies, from speech-recognition specialists to sovereign model builders, that are betting control over the voice stack will be a competitive edge. For CIOs in banking and financial services, the trend points to voice agents that can be tuned for local language, tone and compliance needs instead of being limited by what a third-party API offers.

Arrowhead, which raised $3 million earlier this year in a round led by Stellaris Venture Partners, says it is now extending Fletch’s language coverage for markets beyond India.

Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter.

"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!