Modulate raises $25M for audio-native models that screen deepfake callers

Modulate has raised $25 million in new funding, SiliconANGLE reported on 28 September 2026. Future Ventures led the round, with returning investors Hyperplane and Lakestar participating. The article does not state the round's stage. It puts Modulate's total raised at $60 million including this round, following a $2 million seed and a $30 million Series A in 2022.

The company builds models that analyse raw audio to detect emotion, tone, intent and deepfake signals. Its flagship platform, Velma, uses what the company calls an Ensemble Listening Model architecture combining more than 100 small specialised audio models. Reported customer uses include fraud detection, harassment and grooming detection on gaming and social platforms, screening deepfake callers in healthcare, and evaluating voice AI agents.

Per the article, new SDKs, APIs and industry-specific models are in the works, and the funding will also support hiring and partner integrations. Batch transcription is already available through the existing API at three cents an hour.

Why it matters for digital humans: Two parts of this are directly load-bearing for avatar builders. First, deepfake-caller screening is the defensive mirror of synthetic-voice generation — the same market that sells cloned voices is growing a detection layer, and procurement teams are starting to ask for both. Second, "evaluating voice AI agents" is the quieter gap: most teams shipping conversational avatars still assess them by listening to sample calls, and an audio-native scoring model is a plausible route to measuring prosody, interruption handling and affect at scale rather than by vibe. The architecture claim — 100-plus small specialised models in an ensemble — is the company's own description and has not been independently benchmarked.