HushAI raises $16M to launch privacy-first speech models for consumer apps

AI · 3 min read

HushAI announced a $16 million seed round led by Balderton and launched on-device speech models aimed at messaging apps, voice-note features, and private transcription. The models are quantized for mobile CPUs and promise sub-100ms latency on modern smartphones.

HushAI’s product includes an SDK with local wake-word detection, offline speaker diarization, and encrypted sync for optional cloud features. The company highlights privacy guarantees: audio never leaves the device unless explicitly sent and audited by the user.

Proceeds will be used to optimize model performance for low-power devices, expand language coverage, and build partnerships with consumer messaging platforms. HushAI also announced pilot deals with two encrypted messaging apps.

Privacy-focused design teams welcome tools that provide powerful features without routing audio to the cloud. HushAI’s technical challenge is maintaining model accuracy across noisy environments while keeping latency and power draw minimal.