Back to browse
GitHub Repository

Pure C implementation of FastEnhancer. 3.3x faster than the ONNX Runtime graph, same quality.

0 starsC

Faster-enhancer.c – FastEnhancer speech enhancement, 3.3x faster in C

by kdrkdrkdr·Jul 28, 2026·1 point·0 comments

AI Analysis

●●●BangerWizardryBig Brain

Hand-written int8 GEMM kernels achieve 0.069 RTF on M2 with zero heap allocation.

Strengths
  • Dependency-free C runtime with no inference framework required for deployment.
  • Bit-identical output across NEON, DOTPROD, and I8MM tiers ensures reproducibility.
  • Allocation-free operation after initialization prevents memory jitter in long runs.
Weaknesses
  • Baseline ISA requirement means older hosts fail at initialization without fallback.
  • Niche appeal limited to developers needing extreme audio processing optimization.
Category
Target Audience

Embedded developers, audio engineers, performance-critical applications

Similar To

RNNoise · SpeexDSP · TensorFlow Lite

Similar Projects