Skip to content
Telecom DSP Engineering

EVS Audio Codec Performance.
SIMD-Accelerated Telecom Core.

Standard software implementations of 3GPP Enhanced Voice Services (EVS) choke on high-concurrency telephony gateways. We re-engineered the mathematical DSP pipeline in vectorized AVX-512 and ARM Neon assembly.

Read Case Study

Verified Hardware Benchmarks

Tested on Dual Intel Xeon Platinum 8480+ and Ampere Altra ARM bare-metal nodes.

Throughput
4,800
Concurrent EVS channels per host
CPU Reduction
-78%
Versus 3GPP reference C code
Deterministic Lag
< 12ms
Fixed audio frame processing delay
3GPP Conformance
100%
Bit-exact validation on all ITU vectors

Zero-Copy Linear Resampling

Moroccan carriers route cellular calls over AMR-WB (16kHz) and G.711 (8kHz). Converting between telecommunication sample rates typically introduces audible harmonic artifacts and latency. Our SIMD resampler eliminates dynamic heap allocations, operating directly on pre-allocated ring buffers.

Kernel-Bypass UDP RTP Ingestion

By utilizing DPDK and io_uring, Vext Voice bypasses Linux kernel network stack context switches. Packets are parsed directly in userspace memory, guaranteeing zero packet drops even during 10,000 call concurrency spikes.