Originals · Startup · Startups & Fintech
Arrowhead AI launches Fletch, its own 14 language text to speech model for bank and lender voice agents
The Bengaluru voice AI startup says the model already runs inside a calling stack that handles more than a million customer calls a day.
Bengaluru-based Arrowhead AI has launched Fletch, a text to speech model it built in-house for multilingual customer calls in India and Southeast Asia. The launch was reported on October 30. Fletch covers 14 languages: English, 12 Indian languages including Hindi, Tamil, Telugu and Bengali, plus Malay and Indonesian, and it can switch languages in the middle of a sentence, which is common on Indian calls where Hindi and English mix around an EMI amount or a due date.
The company reports that Fletch starts producing audio in under 100 milliseconds, runs at about 80 milliseconds of latency and offers 12 emotion styles, from calm and empathetic to firm and apologetic. It also claims roughly 30% better pronunciation on selected terms such as brand names, customer names and Indian proper nouns, and is explicit that this figure does not describe overall call resolution or customer satisfaction. All of these numbers are the company's own and have not been independently tested.
Fletch is built on DiTAR, a published approach that pairs a language model with a diffusion transformer to generate continuous speech. Arrowhead says it adapted the architecture and trained it on thousands of hours of curated regional speech plus recordings from professional voice artists. The model sits alongside the company's own speech recognition system, language model and a separate model for reading out addresses, and Arrowhead says the stack powers more than one million calls a day across collections, sales, renewals and inbound service. Named clients include Axis Bank, RBL Bank, IndusInd Bank, Tata 1mg, upGrad, KreditBee, Kissht and TVS Credit.
Arrowhead was founded by Devyani Gupta, a former BCG consultant, and Vengadanathan Srinivasan, an engineer who previously worked at AWS, Rippling, Uber and Airbnb. In January 2026 it raised $3 million in a seed round led by Stellaris Venture Partners, with CRED founder Kunal Shah, M2P founder Madhusudanan R and executives from Turtlemint and Kissht joining as angels.
Owning the voice layer rather than renting it from global speech APIs lets the company tune pronunciation, speed and tone for its own markets. Arrowhead plans to show the Malay and Indonesian versions at the Singapore FinTech Festival on November 18 to 20 as it pushes beyond India.
