Voice AI startup Modulate raises $25M to bring audio-native models to more developers

Voice artificial intelligence startup Modulate Inc. has secured $25 million in new funding to expand its audio-native models for developers. Modulate's AI models analyze raw audio data from conversations, detecting elements such as emotion, tone, intent, and deepfakes. These models can flag potential fraud or caller impatience during live conversations.

Originally focused on moderating voice chat in online games, Modulate's models now assist healthcare institutions in identifying impersonators and help voice AI agents assess performance. The company's flagship platform, Velma, utilizes the Ensemble Listening Model architecture, which efficiently combines results from over 100 specialized audio models.

Modulate reports that over 10 million hours of audio pass through its models each month, exceeding 600 million hours in total. Two of its products have earned first place on Hugging Face leaderboards this year, with a deepfake detection model boasting 98.9% accuracy. Co-founder and CEO Carter Huffman notes that voice is emerging as a primary AI interface, presenting new challenges that require specialized audio intelligence.

Future Ventures led the funding round, with participation from Hyperplane and Lakestar, both of which previously invested in Modulate. The funds will be allocated towards developing industry-specific models, enhancing developer tools, and expanding the team.