Punctuation Restoration

Table of Contents

• Offline Punctuation
• Online Punctuation

sherpa-onnx supports both offline and online punctuation restoration.

See also

SherpaOnnxCreateOfflinePunctuation, SherpaOnnxCreateOnlinePunctuation

Offline Punctuation

Offline punctuation processes the entire input text at once using a CT-Transformer model.

SherpaOnnxOfflinePunctuationConfig config;
memset(&config, 0, sizeof(config));
config.model.ct_transformer =
    "./sherpa-onnx-punct-ct-transformer-zh-en-vocab272727-2024-04-12-int8/model.int8.onnx";
config.model.num_threads = 1;
config.model.provider = "cpu";
 
const SherpaOnnxOfflinePunctuation *punct =
    SherpaOnnxCreateOfflinePunctuation(&config);
 
const char *result = SherpaOfflinePunctuationAddPunct(
    punct, "how are you i am fine thank you");
printf("%s\n", result);
SherpaOfflinePunctuationFreeText(result);
SherpaOnnxDestroyOfflinePunctuation(punct);

Model package: sherpa-onnx-punct-ct-transformer-zh-en-vocab272727-2024-04-12-int8

Example source: add-punctuation-c-api.c

Online Punctuation

Online punctuation processes text incrementally using a CNN-BiLSTM model.

SherpaOnnxOnlinePunctuationConfig config;
memset(&config, 0, sizeof(config));
config.model.cnn_bilstm =
    "./sherpa-onnx-online-punct-en-2024-08-06/model.int8.onnx";
config.model.bpe_vocab =
    "./sherpa-onnx-online-punct-en-2024-08-06/bpe.vocab";
config.model.num_threads = 1;
config.model.provider = "cpu";
 
const SherpaOnnxOnlinePunctuation *punct =
    SherpaOnnxCreateOnlinePunctuation(&config);
 
const char *out = SherpaOnnxOnlinePunctuationAddPunct(
    punct, "how are you i am fine thank you");
printf("%s\n", out);
SherpaOnnxOnlinePunctuationFreeText(out);
SherpaOnnxDestroyOnlinePunctuation(punct);

Model package: sherpa-onnx-online-punct-en-2024-08-06

Example source: add-punctuation-online-c-api.c