Source Separation

Table of Contents

• Spleeter
• UVR (MDX-Net)

sherpa-onnx supports source separation through the SherpaOnnxCreateOfflineSourceSeparation() API. Configure exactly one model family by filling in the corresponding field of SherpaOnnxOfflineSourceSeparationModelConfig.

See also

SherpaOnnxCreateOfflineSourceSeparation, SherpaOnnxOfflineSourceSeparationConfig

Spleeter

Spleeter separates audio into vocals and accompaniment (2 stems).

SherpaOnnxOfflineSourceSeparationConfig config;
memset(&config, 0, sizeof(config));
config.model.spleeter.vocals =
    "./sherpa-onnx-spleeter-2stems-fp16/vocals.fp16.onnx";
config.model.spleeter.accompaniment =
    "./sherpa-onnx-spleeter-2stems-fp16/accompaniment.fp16.onnx";
config.model.num_threads = 1;
config.model.provider = "cpu";
 
const SherpaOnnxOfflineSourceSeparation *ss =
    SherpaOnnxCreateOfflineSourceSeparation(&config);
// ... use ss ...
SherpaOnnxDestroyOfflineSourceSeparation(ss);

Model package: sherpa-onnx-spleeter-2stems-fp16

Example source: source-separation-spleeter-c-api.c

UVR (MDX-Net)

UVR separates audio into vocals and non-vocals.

SherpaOnnxOfflineSourceSeparationConfig config;
memset(&config, 0, sizeof(config));
config.model.uvr.model = "./UVR-MDX-NET-Voc_FT.onnx";
config.model.num_threads = 1;
config.model.provider = "cpu";
 
const SherpaOnnxOfflineSourceSeparation *ss =
    SherpaOnnxCreateOfflineSourceSeparation(&config);
// ... use ss ...
SherpaOnnxDestroyOfflineSourceSeparation(ss);

Model file: UVR-MDX-NET-Voc_FT.onnx

Example source: source-separation-uvr-c-api.c