code refactoring and add CI (#11)

sherpa-onnx/csrc/CMakeLists.txt

@@ -1,13 +1,8 @@
-add_executable(online-fbank-test online-fbank-test.cc)
-target_link_libraries(online-fbank-test kaldi-native-fbank-core)
-
-include_directories( 
-    ${ONNXRUNTIME_ROOTDIR}/include/onnxruntime/core/session/
-    ${ONNXRUNTIME_ROOTDIR}/include/onnxruntime/core/providers/tensorrt/
-)
-
-include_directories(
-    ${KALDINATIVEIO}
-)
+include_directories(${CMAKE_SOURCE_DIR})
 add_executable(sherpa-onnx main.cpp)
-target_link_libraries(sherpa-onnx onnxruntime kaldi-native-fbank-core kaldi_native_io_core)
+
+target_link_libraries(sherpa-onnx
+  onnxruntime
+  kaldi-native-fbank-core
+  kaldi_native_io_core
+)

sherpa-onnx/csrc/main.cpp

@@ -1,101 +1,99 @@
-#include <vector>
-#include <iostream>
 #include <algorithm>
-#include <time.h>
-#include <math.h>
 #include <fstream>
+#include <iostream>
+#include <math.h>
+#include <time.h>
+#include <vector>
 
-#include "fbank_features.h"
-#include "rnnt_beam_search.h"
+#include "sherpa-onnx/csrc/fbank_features.h"
+#include "sherpa-onnx/csrc/rnnt_beam_search.h"
 
 #include "kaldi-native-fbank/csrc/online-feature.h"
 
+int main(int argc, char *argv[]) {
+  char *encoder_path = argv[1];
+  char *decoder_path = argv[2];
+  char *joiner_path = argv[3];
+  char *joiner_encoder_proj_path = argv[4];
+  char *joiner_decoder_proj_path = argv[5];
+  char *token_path = argv[6];
+  std::string search_method = argv[7];
+  char *filename = argv[8];
 
-int main(int argc, char* argv[]) {
-    char* encoder_path = argv[1];
-    char* decoder_path = argv[2];
-    char* joiner_path = argv[3];
-    char* joiner_encoder_proj_path = argv[4];
-    char* joiner_decoder_proj_path = argv[5];
-    char* token_path = argv[6];
-    std::string search_method = argv[7];
-    char* filename = argv[8];
+  // General parameters
+  int numberOfThreads = 16;
 
-    // General parameters
-    int numberOfThreads = 16;
+  // Initialize fbanks
+  knf::FbankOptions opts;
+  opts.frame_opts.dither = 0;
+  opts.frame_opts.samp_freq = 16000;
+  opts.frame_opts.frame_shift_ms = 10.0f;
+  opts.frame_opts.frame_length_ms = 25.0f;
+  opts.mel_opts.num_bins = 80;
+  opts.frame_opts.window_type = "povey";
+  opts.frame_opts.snip_edges = false;
+  knf::OnlineFbank fbank(opts);
 
-    // Initialize fbanks
-    knf::FbankOptions opts;
-    opts.frame_opts.dither = 0;
-    opts.frame_opts.samp_freq = 16000;
-    opts.frame_opts.frame_shift_ms = 10.0f;
-    opts.frame_opts.frame_length_ms = 25.0f;
-    opts.mel_opts.num_bins = 80;
-    opts.frame_opts.window_type = "povey";
-    opts.frame_opts.snip_edges = false;
-    knf::OnlineFbank fbank(opts);
+  // set session opts
+  // https://onnxruntime.ai/docs/performance/tune-performance.html
+  session_options.SetIntraOpNumThreads(numberOfThreads);
+  session_options.SetInterOpNumThreads(numberOfThreads);
+  session_options.SetGraphOptimizationLevel(
+      GraphOptimizationLevel::ORT_ENABLE_EXTENDED);
+  session_options.SetLogSeverityLevel(4);
+  session_options.SetExecutionMode(ExecutionMode::ORT_SEQUENTIAL);
 
-    // set session opts
-    // https://onnxruntime.ai/docs/performance/tune-performance.html
-    session_options.SetIntraOpNumThreads(numberOfThreads);
-    session_options.SetInterOpNumThreads(numberOfThreads);
-    session_options.SetGraphOptimizationLevel(GraphOptimizationLevel::ORT_ENABLE_EXTENDED);
-    session_options.SetLogSeverityLevel(4);
-    session_options.SetExecutionMode(ExecutionMode::ORT_SEQUENTIAL);
-    
-    api.CreateTensorRTProviderOptions(&tensorrt_options);
-    std::unique_ptr<OrtTensorRTProviderOptionsV2, decltype(api.ReleaseTensorRTProviderOptions)> rel_trt_options(tensorrt_options, api.ReleaseTensorRTProviderOptions);
-    api.SessionOptionsAppendExecutionProvider_TensorRT_V2(static_cast<OrtSessionOptions*>(session_options), rel_trt_options.get());
+  api.CreateTensorRTProviderOptions(&tensorrt_options);
+  std::unique_ptr<OrtTensorRTProviderOptionsV2,
+                  decltype(api.ReleaseTensorRTProviderOptions)>
+      rel_trt_options(tensorrt_options, api.ReleaseTensorRTProviderOptions);
+  api.SessionOptionsAppendExecutionProvider_TensorRT_V2(
+      static_cast<OrtSessionOptions *>(session_options), rel_trt_options.get());
 
-    // Define model
-    auto model = get_model(
-        encoder_path,
-        decoder_path,
-        joiner_path,
-        joiner_encoder_proj_path,
-        joiner_decoder_proj_path,
-        token_path
-    );
-    
-    std::vector<std::string> filename_list {
-        filename
-    };
+  // Define model
+  auto model =
+      get_model(encoder_path, decoder_path, joiner_path,
+                joiner_encoder_proj_path, joiner_decoder_proj_path, token_path);
 
-    for (auto filename : filename_list){
-        std::cout << filename << std::endl;
-        auto samples = readWav(filename, true);
-        int numSamples = samples.NumCols();
+  std::vector<std::string> filename_list{filename};
 
-        auto features = ComputeFeatures(fbank, opts, samples);
+  for (auto filename : filename_list) {
+    std::cout << filename << std::endl;
+    auto samples = readWav(filename, true);
+    int numSamples = samples.NumCols();
 
-        auto tic = std::chrono::high_resolution_clock::now();
+    auto features = ComputeFeatures(fbank, opts, samples);
 
-        // # === Encoder Out === #
-        int num_frames = features.size() / opts.mel_opts.num_bins;
-        auto encoder_out = model.encoder_forward(features,
-                                std::vector<int64_t> {num_frames},
-                                std::vector<int64_t> {1, num_frames, 80},
-                                std::vector<int64_t> {1},
-                                memory_info);
+    auto tic = std::chrono::high_resolution_clock::now();
 
-        // # === Search === #
-        std::vector<std::vector<int32_t>> hyps;
-        if (search_method == "greedy")
-            hyps = GreedySearch(&model, &encoder_out);
-        else{
-            std::cout << "wrong search method!" << std::endl;
-            exit(0);
-        }
-        auto results = hyps2result(model.tokens_map, hyps);
+    // # === Encoder Out === #
+    int num_frames = features.size() / opts.mel_opts.num_bins;
+    auto encoder_out =
+        model.encoder_forward(features, std::vector<int64_t>{num_frames},
+                              std::vector<int64_t>{1, num_frames, 80},
+                              std::vector<int64_t>{1}, memory_info);
 
-        // # === Print Elapsed Time === #
-        auto elapsed = std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::high_resolution_clock::now() - tic);
-        std::cout << "Elapsed: " << float(elapsed.count()) / 1000 << " seconds" << std::endl;
-        std::cout << "rtf: " << float(elapsed.count()) / 1000 / (numSamples / 16000)  << std::endl;
-
-        print_hyps(hyps);
-        std::cout << results[0] << std::endl;
+    // # === Search === #
+    std::vector<std::vector<int32_t>> hyps;
+    if (search_method == "greedy")
+      hyps = GreedySearch(&model, &encoder_out);
+    else {
+      std::cout << "wrong search method!" << std::endl;
+      exit(0);
     }
+    auto results = hyps2result(model.tokens_map, hyps);
 
-    return 0;
+    // # === Print Elapsed Time === #
+    auto elapsed = std::chrono::duration_cast<std::chrono::milliseconds>(
+        std::chrono::high_resolution_clock::now() - tic);
+    std::cout << "Elapsed: " << float(elapsed.count()) / 1000 << " seconds"
+              << std::endl;
+    std::cout << "rtf: " << float(elapsed.count()) / 1000 / (numSamples / 16000)
+              << std::endl;
+
+    print_hyps(hyps);
+    std::cout << results[0] << std::endl;
+  }
+
+  return 0;
 }

sherpa-onnx/csrc/rnnt_beam_search.h

@@ -61,7 +61,6 @@ std::vector<std::vector<int32_t>> GreedySearch(
     auto projected_encoder_out = model->joiner_encoder_proj_forward(encoder_out_vector, 
                                         std::vector<int64_t> {encoder_out_dim1, encoder_out_dim2}, 
                                         memory_info);
-
     Ort::Value &projected_encoder_out_tensor = projected_encoder_out[0];
     int projected_encoder_out_dim1 = projected_encoder_out_tensor.GetTensorTypeAndShapeInfo().GetShape()[0];
     int projected_encoder_out_dim2 = projected_encoder_out_tensor.GetTensorTypeAndShapeInfo().GetShape()[1];
@@ -78,12 +77,12 @@ std::vector<std::vector<int32_t>> GreedySearch(
 
         auto logits = model->joiner_forward(cur_encoder_out,
                                             projected_decoder_out_vector,
-                                            std::vector<int64_t> {1, 1, 1, projected_encoder_out_dim2},
-                                            std::vector<int64_t> {1, 1, 1, projected_decoder_out_dim},
+                                            std::vector<int64_t> {1, projected_encoder_out_dim2},
+                                            std::vector<int64_t> {1, projected_decoder_out_dim},
                                             memory_info);
 
         Ort::Value &logits_tensor = logits[0];
-        int logits_dim = logits_tensor.GetTensorTypeAndShapeInfo().GetShape()[3];
+        int logits_dim = logits_tensor.GetTensorTypeAndShapeInfo().GetShape()[1];
         auto logits_vector = ortVal2Vector(logits_tensor, logits_dim);
         
         int max_indices = static_cast<int>(std::distance(logits_vector.begin(), std::max_element(logits_vector.begin(), logits_vector.end())));

sherpa-onnx/csrc/utils_onnx.h

@@ -1,5 +1,5 @@
 #include <iostream>
-#include <onnxruntime_cxx_api.h>
+#include "onnxruntime_cxx_api.h"
 
 Ort::Env env(ORT_LOGGING_LEVEL_WARNING, "test");
 const auto& api = Ort::GetApi();