Gradium

Gradium is a voice AI base for developers, covering four capabilities: text-to-speech, speech-to-text, real-time translation and voice cloning, focusing on ultra-low latency.

Its target scenario is a voice agent—the kind of AI that can answer calls and have real-time conversations. This type of application is extremely sensitive to delays. The threshold for pauses that people can perceive in a conversation is very low. The difference of a few hundred milliseconds directly determines whether the other party feels uncomfortable. Gradium focuses optimization on this link, rather than simply focusing on sound quality.

The company has received US$100 million in seed round expansion financing, including NVIDIA, and has set up an office in the San Francisco Bay Area. This background is critical for companies building real-time inference infrastructure, because the hardware cost of this road is very heavy.

The product is provided in the form of API, and the official website has a demo module for online listening, supporting multiple languages. It is suitable for development teams that want to make voice assistants, customer service robots, real-time subtitles or dubbing products. There are not many scenarios where individual users can directly use it.