EngineX-Iluvatar/enginex-mr_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Reduce model initialization time for offline speech recognition (#213)

Browse Source
This commit is contained in:
Fangjun Kuang
2023-07-14 18:07:27 +08:00
committed by GitHub
parent 0abd7ce881
commit f3206c49dc
10 changed files with 109 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

63
sherpa-onnx/c-api/c-api.cc
View File

@@ -33,23 +33,33 @@ SherpaOnnxOnlineRecognizer *CreateOnlineRecognizer(
const SherpaOnnxOnlineRecognizerConfig *config) {
sherpa_onnx::OnlineRecognizerConfig recognizer_config;
recognizer_config.feat_config.sampling_rate = SHERPA_ONNX_OR(config->feat_config.sample_rate, 16000);
recognizer_config.feat_config.feature_dim = SHERPA_ONNX_OR(config->feat_config.feature_dim, 80);
recognizer_config.feat_config.sampling_rate =
SHERPA_ONNX_OR(config->feat_config.sample_rate, 16000);
recognizer_config.feat_config.feature_dim =
SHERPA_ONNX_OR(config->feat_config.feature_dim, 80);
recognizer_config.model_config.encoder_filename =
SHERPA_ONNX_OR(config->model_config.encoder, "");
recognizer_config.model_config.decoder_filename =
SHERPA_ONNX_OR(config->model_config.decoder, "");
recognizer_config.model_config.joiner_filename = SHERPA_ONNX_OR(config->model_config.joiner, "");
recognizer_config.model_config.tokens = SHERPA_ONNX_OR(config->model_config.tokens, "");
recognizer_config.model_config.num_threads = SHERPA_ONNX_OR(config->model_config.num_threads, 1);
recognizer_config.model_config.provider = SHERPA_ONNX_OR(config->model_config.provider, "cpu");
recognizer_config.model_config.debug = SHERPA_ONNX_OR(config->model_config.debug, 0);
recognizer_config.model_config.joiner_filename =
SHERPA_ONNX_OR(config->model_config.joiner, "");
recognizer_config.model_config.tokens =
SHERPA_ONNX_OR(config->model_config.tokens, "");
recognizer_config.model_config.num_threads =
SHERPA_ONNX_OR(config->model_config.num_threads, 1);
recognizer_config.model_config.provider =
SHERPA_ONNX_OR(config->model_config.provider, "cpu");
recognizer_config.model_config.debug =
SHERPA_ONNX_OR(config->model_config.debug, 0);
recognizer_config.decoding_method = SHERPA_ONNX_OR(config->decoding_method, "greedy_search");
recognizer_config.max_active_paths = SHERPA_ONNX_OR(config->max_active_paths, 4);
recognizer_config.decoding_method =
SHERPA_ONNX_OR(config->decoding_method, "greedy_search");
recognizer_config.max_active_paths =
SHERPA_ONNX_OR(config->max_active_paths, 4);
recognizer_config.enable_endpoint = SHERPA_ONNX_OR(config->enable_endpoint, 0);
recognizer_config.enable_endpoint =
SHERPA_ONNX_OR(config->enable_endpoint, 0);
recognizer_config.endpoint_config.rule1.min_trailing_silence =
SHERPA_ONNX_OR(config->rule1_min_trailing_silence, 2.4);
@@ -173,9 +183,11 @@ SherpaOnnxOfflineRecognizer *CreateOfflineRecognizer(
const SherpaOnnxOfflineRecognizerConfig *config) {
sherpa_onnx::OfflineRecognizerConfig recognizer_config;
recognizer_config.feat_config.sampling_rate = SHERPA_ONNX_OR(config->feat_config.sample_rate, 16000);
recognizer_config.feat_config.sampling_rate =
SHERPA_ONNX_OR(config->feat_config.sample_rate, 16000);
recognizer_config.feat_config.feature_dim = SHERPA_ONNX_OR(config->feat_config.feature_dim, 80);
recognizer_config.feat_config.feature_dim =
SHERPA_ONNX_OR(config->feat_config.feature_dim, 80);
recognizer_config.model_config.transducer.encoder_filename =
SHERPA_ONNX_OR(config->model_config.transducer.encoder, "");
@@ -184,7 +196,7 @@ SherpaOnnxOfflineRecognizer *CreateOfflineRecognizer(
SHERPA_ONNX_OR(config->model_config.transducer.decoder, "");
recognizer_config.model_config.transducer.joiner_filename =
SHERPA_ONNX_OR(config->model_config.transducer.joiner,"");
SHERPA_ONNX_OR(config->model_config.transducer.joiner, "");
recognizer_config.model_config.paraformer.model =
SHERPA_ONNX_OR(config->model_config.paraformer.model, "");
@@ -192,15 +204,26 @@ SherpaOnnxOfflineRecognizer *CreateOfflineRecognizer(
recognizer_config.model_config.nemo_ctc.model =
SHERPA_ONNX_OR(config->model_config.nemo_ctc.model, "");
recognizer_config.model_config.tokens = SHERPA_ONNX_OR(config->model_config.tokens, "");
recognizer_config.model_config.num_threads = SHERPA_ONNX_OR(config->model_config.num_threads, 1);
recognizer_config.model_config.debug = SHERPA_ONNX_OR(config->model_config.debug, 0);
recognizer_config.model_config.tokens =
SHERPA_ONNX_OR(config->model_config.tokens, "");
recognizer_config.model_config.num_threads =
SHERPA_ONNX_OR(config->model_config.num_threads, 1);
recognizer_config.model_config.debug =
SHERPA_ONNX_OR(config->model_config.debug, 0);
recognizer_config.model_config.provider =
SHERPA_ONNX_OR(config->model_config.provider, "cpu");
recognizer_config.model_config.model_type =
SHERPA_ONNX_OR(config->model_config.model_type, "");
recognizer_config.lm_config.model = SHERPA_ONNX_OR(config->lm_config.model, "");
recognizer_config.lm_config.scale = SHERPA_ONNX_OR(config->lm_config.scale, 1.0);
recognizer_config.lm_config.model =
SHERPA_ONNX_OR(config->lm_config.model, "");
recognizer_config.lm_config.scale =
SHERPA_ONNX_OR(config->lm_config.scale, 1.0);
recognizer_config.decoding_method = SHERPA_ONNX_OR(config->decoding_method, "greedy_search");
recognizer_config.max_active_paths = SHERPA_ONNX_OR(config->max_active_paths, 4);
recognizer_config.decoding_method =
SHERPA_ONNX_OR(config->decoding_method, "greedy_search");
recognizer_config.max_active_paths =
SHERPA_ONNX_OR(config->max_active_paths, 4);
if (config->model_config.debug) {
fprintf(stderr, "%s\n", recognizer_config.ToString().c_str());