Recently, smallest.ai, an AI startup headquartered in San Francisco, California, launched its new product, Lightning, a text-to-speech (TTS) model capable of generating 10-second audio clips in just 100 milliseconds. This technological advancement allows global developers to build highly realistic voicebot applications with extremely low latency, reducing implementation costs and enhancing accessibility.

image.png

Lightning currently supports various accents in English and Hindi, with the team planning to quickly add more languages to meet market demands. Priced at only $0.02 per minute (approximately 1.6 Indian Rupees), this model offers a highly cost-effective solution for voicebot developers, keeping operational costs below 1 Rupee per minute and significantly reducing the expenses associated with building voicebots, while expanding market reach.

Unlike traditional TTS models that rely on streaming and web sockets, increasing server burdens and complex scalability, Lightning delivers audio via a simple REST API design within approximately 100 milliseconds, avoiding server strain from continuous streaming. This rapid processing capability and cost efficiency make it a notable alternative in the voicebot industry.

smallest.ai was founded by alumni of the Indian Institute of Technology Guwahati, Sudarshan Kamath and Akshat Mandloi. Kamath stated that smallest.ai's low-cost strategy is due to their focus on data quality and model efficiency. "Our model is much smaller than competitors like ElevenLabs, but we achieve high-quality voice output through highly refined data," he explained.

Early adopters of Lightning reported an eightfold reduction in operational costs while experiencing improved audio quality. Beyond real-time voicebot applications, Lightning can be used for creating audiobooks and voiceovers for social media content on platforms like Instagram and YouTube. Non-developers can also access Lightning through the Waves Speech platform, experiencing features such as voice cloning and accent transformation, which are currently in beta testing.

Kamath shared with Analytics India Magazine in an exclusive interaction, "When we started building, we realized that existing voicebot models for Indian languages were not mature enough. The existing models for non-English languages simply couldn't meet production requirements."

In June of this year, smallest.ai also introduced the AWAAZ model, which supports voice cloning through short audio clips at competitive prices. The model aims to meet the scalable applications in regional language markets and provides enterprise-level security and compliance. When asked about their mission, Kamath said, "Why aren't a billion people interacting with AI voices daily, despite the significant advancements in voice AI technology? That's the problem we're striving to solve."

Project entry: https://smallest.ai/blog/lightning-fast-text-to-speech

Key Points:

🌟 The Lightning text-to-speech model generates audio in 100 milliseconds, supporting various accents in English and Hindi, with plans to expand to more languages.

💰 At just $0.02 per minute, it significantly reduces operational costs for voicebot developers.

📱 Lightning is not only suitable for voicebots but also for audiobooks and social media voiceovers, making it accessible for both developers and non-developers.