Silero VAD NCNN

Convert the silero-vad v4 to NCNN, only 16K is supported.

1. vad/collections/resources/silero.jit the original jit model

2. vad/collections/resources/silero_vad.jit the generated jit model.

3. vad/collections/resources/silero.ncnn.param the ncnn model.

Quick Start

• main.py demo scripts.

• convert.py Conversion script to generate 'silero.jit', which also supports ncnn model by pnnx, but there are some issues, pnnx==20240819 fixed. Generate silero.jit and then convert it via the old pnnx version. pnnx-20231010-ubuntu.zip work!!!

  ./pnnx silero.jit inputshape=[1,512],[2,1,64],[2,1,64]

• compare.py Compare the original jit model with the generated jit model and the ncnn model.

Cpp implementation

• implementtion silero-vad in ncnn cpp code.
• please modify the ncnn include/link path in CMakeLists.txt first
• build step as follows:

    cd vad-cpp
    mkdir build && cd build
    cmake ..
    make -j4