IIT graduates-led 'Smallest.ai' unveils World's fastest real-time Text to Speech model
The world's fastest real-time text-to-speech model, lightning, has been launched by Smallest.ai, a top developer of multi-modal AI based models with its headquarters located in San Francisco, California. Lightning can produce up to 10 seconds of audio in just 100 milliseconds. This significantly streamlines the integration process while allowing voicebot providers worldwide to create incredibly lifelike bots with sub-second latency.
With prices starting at 0.02 USD/min (1.6 Rs/min), Lightning is also significantly less expensive than its Western rivals, which is another factor that makes it revolutionary. This makes it possible to employ voicebots on a population scale for less than 1 rupee a minute.
Real-time text-to-speech models typically need streaming, which creates a web socket connection. This raises the server's processing burden and makes scaling AI-based tools and voicebots more challenging and costly. Lightning makes it feasible to obtain audio using a basic REST API in about 100ms, which allows bot providers to grow more quickly while significantly reducing API expenses.
Several Hindi and English accents are presently supported by Lightning. In the upcoming months, Smallest.ai intends to integrate additional Asian, European, and Indian languages.
Voicebot platforms that had early access to Lightning saw a significant improvement in speech quality and an eight-fold reduction in their cost per minute.
Although Lightning is designed for real-time applications, it can also be used to provide voiceovers for reels on YouTube, Instagram, and several other social media sites, as well as audiobooks. Lightning is accessible to non-developers via the Waves Speech platform, which offers beta versions of services like accent conversion and voice cloning.
Sudarshan Kamath, Founder of Smallest.ai says, "Why are 1B humans not speaking to AI voices everyday despite incredible advancements in Voice AI? This is the problem we are trying to solve."
