Scaled Cognition launches Voice at $9 for each million characters

Scaled Cognition, a vendor of customer support AI, launched Voice. Voice is a text-to-speech model: it changes written text to spoken audio. The company says it reads account numbers, addresses and other information more correctly than rival services, and at a lower cost. Hosted use costs $9 for each million characters, and self-hosting costs $1.

How it works

Voice finds the meaning of a sentence before it speaks. For example, it reads “Pay 1/2 by 1/2” as “pay half by January second”. It has more than 25 languages and voice cloning, which copies the voice of a person. Firms use it through an API (a software connection) or on their own servers.

The numbers

Self-hosting costs $1 for each million characters, without servers. The company estimates about $13 with servers at high use. It says other hosted voices cost $35 to $100. It claims 70 milliseconds from text to the start of voice.

What is not known

The company offers a blind listening test. First feedback favored Voice, but the company gave no sample sizes or results. At low use, self-hosting can cost more than the hosted price. The company did not name the rivals in its price claim.

Insight: Rivals show scores and this launch shows none

Artificial Analysis, a test group, ranks Fish Audio Drama 3 Preview at 93.2% for pronunciation, more than Eleven v4 at 91.7%. Scaled Cognition gives a price and a blind test with no results. This looks like a fight on price and on correct voice output, with no public score at this time. Run your own test before you buy.