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

Code refactoring (#74)

Browse Source 
* Don't reset model state and feature extractor on endpointing

* support passing decoding_method from commandline

* Add modified_beam_search to Python API

* fix C API example

* Fix style issues
This commit is contained in:
Fangjun Kuang
2023-03-03 12:10:59 +08:00
committed by GitHub
parent c241f93c40
commit 7f72c13d9a
34 changed files with 744 additions and 374 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
sherpa-onnx/csrc/sherpa-onnx.cc
View File

@@ -14,7 +14,7 @@
#include "sherpa-onnx/csrc/wave-reader.h"
int main(int32_t argc, char *argv[]) {
if (argc < 6 || argc > 7) {
if (argc < 6 || argc > 8) {
const char *usage = R"usage(
Usage:
./bin/sherpa-onnx \
@@ -22,7 +22,10 @@ Usage:
/path/to/encoder.onnx \
/path/to/decoder.onnx \
/path/to/joiner.onnx \
/path/to/foo.wav [num_threads]
/path/to/foo.wav [num_threads [decoding_method]]
Default value for num_threads is 2.
Valid values for decoding_method: greedy_search (default), modified_beam_search.
Please refer to
https://k2-fsa.github.io/sherpa/onnx/pretrained_models/index.html
@@ -45,9 +48,15 @@ for a list of pre-trained models to download.
std::string wav_filename = argv[5];
config.model_config.num_threads = 2;
if (argc == 7) {
if (argc == 7 && atoi(argv[6]) > 0) {
config.model_config.num_threads = atoi(argv[6]);
}
if (argc == 8) {
config.decoding_method = argv[7];
}
config.max_active_paths = 4;
fprintf(stderr, "%s\n", config.ToString().c_str());
sherpa_onnx::OnlineRecognizer recognizer(config);
@@ -98,6 +107,7 @@ for a list of pre-trained models to download.
1000.;
fprintf(stderr, "num threads: %d\n", config.model_config.num_threads);
fprintf(stderr, "decoding method: %s\n", config.decoding_method.c_str());
fprintf(stderr, "Elapsed seconds: %.3f s\n", elapsed_seconds);
float rtf = elapsed_seconds / duration;