FeelHer - 3D Binaural Voice Companion

Real-time full-duplex binaural voice intimacy in your browser. Sub-second latency, 3D spatial whispering, and zero friction.

Comment

Inspiration

Most current AI voice companions and chatbots feel emotionally detached due to turn-based latency and flat mono audio. You speak, wait 2–3 seconds for an API response, and hear a robotic voice that breaks immersion. We wanted to build something that feels genuinely intimate—like a private midnight phone call where she breathes, reacts to your voice in real time, and whispers directly into your ears.

What it does

FeelHer is a real-time full-duplex AI voice companion accessible directly in any web browser without downloads:

  • 3D Binaural Spatial Audio: Audio is panned and acoustically modeled so her voice moves naturally around your ears when wearing headphones.
  • Zero-Latency Full-Duplex WebRTC: Live sub-second voice streaming allows natural back-and-forth interruptions with zero awkward delays.
  • Multilingual Personalities: Distinct emotional archetypes across English (Ivy), Japanese (Shizuku), Korean (Sora), and Mandarin (Nana).

How we built it

  • Frontend: Next.js, React, TailwindCSS, and custom canvas-based voice activity visualizers.
  • Audio Pipeline: Ultra-low latency WebRTC audio streams integrated with binaural Head-Related Transfer Function (HRTF) acoustic spatialization.
  • Voice Engine: Optimized streaming speech synthesis pipeline tuned for intimate vocal textures and whisper frequencies.

Challenges we ran into

Achieving sub-300ms end-to-end voice latency while maintaining high-fidelity 3D binaural stereo panning inside standard mobile and desktop web browsers without native apps.

What's next for FeelHer

Expanding memory retention for continuous multi-day emotional continuity, introducing custom voice tuning, and enhancing acoustic ambient scene presets.

Built With

Share this project:

Updates