Realtime TTS-2

Realtime TTS-2 is a text-to-speech program made by Inworld AI that turns written text into a human-sounding voice.

Most programs like this speak in only one preset tone, but this one can change tone and emotion in real time based on instructions like "sound brighter" or "a little sadder," almost like directing an actor's performance. It can also switch between more than 500 dialects while keeping the same voice identity, and according to the company, this processing happens within 150 milliseconds, making it flow as naturally as a real conversation.

It's a foundational technology that can be used in various services where a computer needs to speak in a human voice, such as call center bots or audiobook narration.

In the article, it says "Inworld AI officially launched Realtime TTS-2 and took the top spot on the Artificial Analysis leaderboard." However, this No. 1 ranking is based on the company's own citation of results measured by Artificial Analysis, an independent evaluator that scores various voice models—and it's worth noting that rankings shift often, with Cartesia's Sonic-3.6 having held the top spot on the same leaderboard just two weeks earlier.

If you can try a text-to-speech service that lets you adjust tone through instructions like Realtime TTS-2, compare the results by changing only the instruction for the same sentence.

1. Pick one short sentence. Example: "Today's meeting starts at 3 PM."
2. Have it read with Instruction A: "Read this in a calm, businesslike tone."
3. Have it read the same sentence again with Instruction B: "Read this in an excited, upbeat tone."
4. Compare how differently the same words sound depending on the instruction.