Scaled Cognition launches Voice at $9 per million characters

Scaled Cognition, which raised $100 million in Series A funding led by Khosla Ventures, is positioning speech as another component of its enterprise AI platform. The company already offers APT, a model for customer-service conversations and actions in enterprise systems, and says pairing APT with Voice supplies conversation context that can help resolve ambiguities beyond a single sentence.

The launch targets a gap between natural-sounding speech and AI that reliably communicates information customers need to complete transactions. Akshay Kannan, vice president of product management, said modern generative TTS models often stumble over details like reading back an alphanumeric confirmation code or a street address, and that these difficulties become more pronounced in languages other than English. Voice's Smart Pronunciation capability works out what a sentence means before saying a word, the company said.

On cost, Kannan estimated that a redundant setup using two GPUs on standard cloud hardware costs roughly $6,000 a month and supports about 500 million characters, working out to about $13 per million characters including infrastructure, and less than $7 at 1.5 billion characters. He compared that to between $35 and $100 per million characters charged by other hosted voices, which he declined to name. For enterprises that already run their own infrastructure, Voice is designed to drop in, and its compatibility could offer a way to update interactive voice response systems without replacing underlying applications.