PreVox™ Neural RT-20 is a real-time neural noise cancellation and speech enhancement solution for embedded systems, designed as a pre-processor for vocoders. It delivers maximum speech quality while maintaining real-time latency, ensuring cleaner input for any voice codec.
- Hybrid streaming neural core (dual-branch: spectral + time-domain).
- Vocoding pre-processor: improves codec efficiency by feeding it with cleaner speech.
- Low latency. Typically 20–40 ms algorithmic delay (configurable).
- Higher quality. Cleaner speech in stationary & non-stationary noise; no “musical noise,” natural prosody retained.
- Customer retraining. We adapt the model to your noise profiles and device acoustics.
- Optimized for ARM & DSP. Runs on ARM (Cortex-A/M/R, NEON) and common DSP cores (e.g., TI).
Please contact us for details.