Neural and DSP voice processing for modern communication

PreVox™ Suite is our family of voice-processing modules for radio, VoIP and embedded systems. The modules combine neural and classic DSP approaches to deliver clear, intelligible speech and reliable signaling under real-world conditions.

Modules:

• PreVox™ Neural RT — Real-time noise suppression and speech enhancement (RT-5/RT-10/RT-20). 
  Designed as a vocoder pre-processor to maximize intelligibility at low latency.
• PreVox™ Neural NRT — Offline/full-context enhancement for the highest speech quality when real-time is not required.
• PreVox™ Classic NS — DSP-based noise suppressors for resource-constrained targets.
• PreVox™ G.168 LEC — ITU-T-compliant line echo canceller (Lite / Full).
• PreVox™ AGC — Automatic gain control with ultra-low distortion and fast anti-clipping response.
• PreVox™ VAD — Voice activity detector that operates without noise-specific tuning; typically used inside vocoders, but can run standalone.
• PreVox™ Tone Relay — In-band tone transport: detects single/dual tones, inserts compact codes, and phase-stitches tones on receive — artifact-free, even inside speech.

Typical Signal Chain

Neural RT / Classic NS → AGC → VAD → Vocoder (DTX/CNG integrated) → Tone Relay → G.168 LEC (network side)

Why PreVox™

• Vocoder-ready: modules are built as front-end/pre-processing for modern voice codecs.
• Low latency options: choose RT-5 / RT-10 / RT-20 to balance quality vs. delay.
• Robust to real noise: stationary and non-stationary conditions; reverberation handling.
• Embedded-friendly: implementations for DSP/embedded targets; desktop builds for integration and testing.
• One ecosystem: modules interoperate out of the box for faster integration.

Please contact us for details.