smallest.ai has unveiled the beta version of AWAAZ, a text-to-speech (TTS) model designed for Indian languages. This model boasts several advanced features including state-of-the-art Mean Opinion Score (MOS) in Hindi and Indian English, and the ability to speak in over 10 accents.
AWAAZ offers single-shot voice cloning from just a 5-second audio clip and provides a low 200ms streaming latency. In an enticing introductory offer, it is priced at Rs. 999 for 500,000 characters, which it claims is ten times cheaper than its competitors.
smallest.ai developed AWAAZ in response to the lack of high-quality and cost-effective TTS models for Indian languages. The team noted that existing models either suffer from poor quality or are prohibitively expensive, especially for scaling. AWAAZ addresses these issues by utilising high-quality, multi-language, multi-accent datasets, focusing initially on India and South Asia.
A demo showcasing AWAAZ’s capabilities is available, and smallest.ai is seeking feedback from users. Key features of AWAAZ include ~200ms latency, dedicated throughput, enterprise security, custom compliance, and enterprise discounts.
With a headquarters in San Francisco, smallest.ai was founded in 2023 by Sudarshan Kamath and Akshat Mandloi and came out of stealth mode just recently with the aim of building voice-based generative AI models.
The post smallest.ai Launches AWAAZ, a Multi-Lingual, Multi-Accent Text-to-Speech Model in Indian Languages appeared first on AIM.
