A voice profile is the combination of characteristics that makes a voice distinct, from tone and pitch to accent, cadence, and speaking style. These characteristics shape how your voice agent is perceived and how people respond to it.
Today, voice agent companies often choose voices by having a small group of people listen to samples and pick what sounds best. But what sounds best to your team may not be what your users prefer.
Different voices work better for different use cases. Sona Library lets voice agent companies compare real human voices in context, see which ones people prefer, and license the voices they want to clone for their agents.
Listeners hear two voices answer the same prompt, blind, and choose the person they would rather keep talking to. Scores come from the Bradley-Terry model over those comparisons; the ± is the 95% interval around each one, so two voices whose intervals overlap are not distinguishable. Open a voice to hear every prompt it answered.
The two prompts are ranked separately, because they are not the same skill: explaining a product to a customer and walking a new colleague through a task reward different voices. Overall pools the comparisons from both.
A voice needs 50 comparisons on a board before it is ranked there. Below that the interval is too wide to place it, so it is listed as unranked rather than ranked low.
Liva AI