There is a moment in almost every voice AI project where someone has to sit down and listen to dozens of synthetic voices, one after another, trying to decide which one sounds right for the product. It is tedious, subjective, and often done under time pressure. The result is frequently good enough rather than genuinely right.
Together AI just built a tool to fix that. Their new voice search catalog lets teams describe what they need in plain English, upload a reference audio clip if they have one, and get ranked results from over 600 voices across providers like Cartesia, Deepgram, MiniMax, and Rime. You can filter by accent, language, age, emotional tone, and speaking style, and listen to every option directly in the browser.
The company behind this has real weight. Together AI raised $305 million in early 2025 at a $3.3 billion valuation, with backing from NVIDIA, Salesforce Ventures, and Kleiner Perkins. More recently, reports suggest they are now generating close to $1 billion in annualized revenue, making them one of the faster-growing infrastructure companies in AI. Voice is a relatively new expansion for them, but it fits logically into a platform that already handles the full chain of building AI applications.
The real story here is not the tool. It is what the tool is responding to.
Voice agents are expanding fast. The market was worth around $2.5 billion this year and is projected to reach over $35 billion by 2033, driven by customer support automation in banking, insurance, retail, and healthcare. Customer support alone already accounts for nearly half of all voice agent deployments. These are not internal tools. They are the first voice your customer hears when they call your company.
The voice an agent uses shapes how the customer feels about the entire interaction. Research on voice persuasion shows that specific voice attributes, including tone, age, and gender, directly influence consumer decisions and brand trust. One study found that consistent voice identity across channels improves how customers perceive both trust and competence. A financial services firm and a youth-facing travel brand do not sound the same for good reason.
The problem has been that picking the wrong voice carries real cost. When automated voice agents fail to feel right, customers notice. Some research shows that a bad automated voice experience pushes roughly a quarter of customers toward competitors and, in the worst cases, makes 13 percent decide never to do business with that company again. Yet the tools for making a good voice choice have barely existed. Provider catalogs list voices alphabetically, documentation rarely tells you what the voice is suited for, and teams have had to rely on gut instinct and manual listening sessions.
What Together AI has done is treat voice selection as a data problem. Every voice in their catalog has been tagged across over 15 attributes by an AI that has listened to all of them. That metadata powers both the search and the filtering. The result is something closer to how a casting director thinks about voice, matching characteristics to context rather than picking from an arbitrary list.
This also points to a deeper shift. The voice AI space is becoming genuinely competitive and fragmented. ElevenLabs, Cartesia, Deepgram, PlayHT, and dozens of others are all competing on quality, latency, and price. A platform that aggregates them and helps teams compare across providers has a natural advantage, because no single provider will win on every dimension for every use case.
For businesses outside the tech industry, the takeaway is this: if you are planning to deploy a voice agent, or already have one, the voice it uses deserves the same deliberate attention as your logo or your customer service scripts. It is speaking for your company, and customers are already forming opinions the moment it says hello.