Audio Tagging

Table of Contents

• Zipformer
• CED

sherpa-onnx supports audio tagging through the SherpaOnnxCreateAudioTagging() API. Configure exactly one model family by filling in the corresponding field of SherpaOnnxAudioTaggingModelConfig.

See also

SherpaOnnxCreateAudioTagging, SherpaOnnxAudioTaggingConfig

Zipformer

The Zipformer audio tagging model uses a CSV file for class labels.

SherpaOnnxAudioTaggingConfig config;
memset(&config, 0, sizeof(config));
config.model.zipformer.model =
    "./sherpa-onnx-zipformer-audio-tagging-2024-04-09/model.int8.onnx";
config.model.num_threads = 1;
config.model.provider = "cpu";
config.labels =
    "./sherpa-onnx-zipformer-audio-tagging-2024-04-09/class_labels_indices.csv";
config.top_k = 5;
 
const SherpaOnnxAudioTagging *tagger = SherpaOnnxCreateAudioTagging(&config);

Model package: sherpa-onnx-zipformer-audio-tagging-2024-04-09

Example source: audio-tagging-c-api.c

CED

CED (Conditional Event Detection) is an alternative audio tagging model.

SherpaOnnxAudioTaggingConfig config;
memset(&config, 0, sizeof(config));
config.model.ced = "./sherpa-onnx-audio-tagging-ced-mini-int8/model.int8.onnx";
config.model.num_threads = 1;
config.model.provider = "cpu";
config.top_k = 5;
 
const SherpaOnnxAudioTagging *tagger = SherpaOnnxCreateAudioTagging(&config);

Model package: sherpa-onnx-audio-tagging-ced-mini-int8