From 94de39dde2e616a01683c518023d0fab72b4e103 Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期一, 19 二月 2024 22:21:50 +0800
Subject: [PATCH] aishell example

---
 runtime/onnxruntime/src/audio.cpp |   75 +++++++++++++++++++++++++++++++++++++
 1 files changed, 75 insertions(+), 0 deletions(-)

diff --git a/runtime/onnxruntime/src/audio.cpp b/runtime/onnxruntime/src/audio.cpp
index 40ea871..9b93dc8 100644
--- a/runtime/onnxruntime/src/audio.cpp
+++ b/runtime/onnxruntime/src/audio.cpp
@@ -133,6 +133,7 @@
     };
     ~AudioWindow(){
         free(window);
+        window = nullptr;
     };
     int put(int val)
     {
@@ -162,6 +163,7 @@
 AudioFrame::~AudioFrame(){
     if(data != nullptr){
         free(data);
+        data = nullptr;
     }
 }
 int AudioFrame::SetStart(int val)
@@ -221,12 +223,15 @@
 {
     if (speech_buff != nullptr) {
         free(speech_buff);
+        speech_buff = nullptr;
     }
     if (speech_data != nullptr) {
         free(speech_data);
+        speech_data = nullptr;
     }
     if (speech_char != nullptr) {
         free(speech_char);
+        speech_char = nullptr;
     }
     ClearQueue(frame_queue);
     ClearQueue(asr_online_queue);
@@ -271,6 +276,7 @@
     speech_len = samples.size();
     if (speech_data != nullptr) {
         free(speech_data);
+        speech_data = nullptr;
     }
     speech_data = (float*)malloc(sizeof(float) * speech_len);
     memset(speech_data, 0, sizeof(float) * speech_len);
@@ -402,9 +408,11 @@
 
     if (speech_data != nullptr) {
         free(speech_data);
+        speech_data = nullptr;
     }
     if (speech_char != nullptr) {
         free(speech_char);
+        speech_char = nullptr;
     }
     offset = 0;
     
@@ -584,6 +592,7 @@
 
     if (speech_data != nullptr) {
         free(speech_data);
+        speech_data = nullptr;
     }
 
     speech_len = (resampled_buffers.size()) / 2;
@@ -615,9 +624,11 @@
     WaveHeader header;
     if (speech_data != nullptr) {
         free(speech_data);
+        speech_data = nullptr;
     }
     if (speech_buff != nullptr) {
         free(speech_buff);
+        speech_buff = nullptr;
     }
     
     offset = 0;
@@ -690,6 +701,7 @@
     WaveHeader header;
     if (speech_char != nullptr) {
         free(speech_char);
+        speech_char = nullptr;
     }
     offset = 0;
     std::ifstream is(filename, std::ifstream::binary);
@@ -729,9 +741,11 @@
     WaveHeader header;
     if (speech_data != nullptr) {
         free(speech_data);
+        speech_data = nullptr;
     }
     if (speech_buff != nullptr) {
         free(speech_buff);
+        speech_buff = nullptr;
     }
 
     std::memcpy(&header, buf, sizeof(header));
@@ -774,6 +788,7 @@
 {
     if (speech_data != nullptr) {
         free(speech_data);
+        speech_data = nullptr;
     }
 
     speech_len = n_buf_len / 2;
@@ -807,6 +822,7 @@
 {
     if (speech_data != nullptr) {
         free(speech_data);
+        speech_data = nullptr;
     }
 
     speech_len = n_buf_len / 2;
@@ -844,9 +860,11 @@
 {
     if (speech_data != nullptr) {
         free(speech_data);
+        speech_data = nullptr;
     }
     if (speech_buff != nullptr) {
         free(speech_buff);
+        speech_buff = nullptr;
     }
     offset = 0;
 
@@ -899,6 +917,7 @@
 {
     if (speech_char != nullptr) {
         free(speech_char);
+        speech_char = nullptr;
     }
     offset = 0;
 
@@ -926,6 +945,7 @@
 {
     if (speech_char != nullptr) {
         free(speech_char);
+        speech_char = nullptr;
     }
 
     FILE* fp;
@@ -1030,6 +1050,7 @@
         new_data[tmp_off + i] = speech_data[ii];
     }
     free(speech_data);
+    speech_data = nullptr;
     speech_data = new_data;
     speech_len = num_new_samples;
 
@@ -1064,6 +1085,60 @@
     }
 }
 
+void Audio::CutSplit(OfflineStream* offline_stream)
+{
+    std::unique_ptr<VadModel> vad_online_handle = make_unique<FsmnVadOnline>((FsmnVad*)(offline_stream->vad_handle).get());
+    AudioFrame *frame;
+
+    frame = frame_queue.front();
+    frame_queue.pop();
+    int sp_len = frame->GetLen();
+    delete frame;
+    frame = nullptr;
+
+    int step = dest_sample_rate*10;
+    bool is_final=false;
+    vector<std::vector<int>> vad_segments;
+    for (int sample_offset = 0; sample_offset < speech_len; sample_offset += std::min(step, speech_len - sample_offset)) {
+        if (sample_offset + step >= speech_len - 1) {
+                step = speech_len - sample_offset;
+                is_final = true;
+            } else {
+                is_final = false;
+        }
+        std::vector<float> pcm_data(speech_data+sample_offset, speech_data+sample_offset+step);
+        vector<std::vector<int>> cut_segments = vad_online_handle->Infer(pcm_data, is_final);
+        vad_segments.insert(vad_segments.end(), cut_segments.begin(), cut_segments.end());
+    }    
+
+    int speech_start_i = -1, speech_end_i =-1;
+    for(vector<int> vad_segment:vad_segments)
+    {
+        if(vad_segment.size() != 2){
+            LOG(ERROR) << "Size of vad_segment is not 2.";
+            break;
+        }
+        if(vad_segment[0] != -1){
+            speech_start_i = vad_segment[0];
+        }
+        if(vad_segment[1] != -1){
+            speech_end_i = vad_segment[1];
+        }
+
+        if(speech_start_i!=-1 && speech_end_i!=-1){
+            frame = new AudioFrame();
+            int start = speech_start_i*seg_sample;
+            int end = speech_end_i*seg_sample;
+            frame->SetStart(start);
+            frame->SetEnd(end);
+            frame_queue.push(frame);
+            frame = nullptr;
+            speech_start_i=-1;
+            speech_end_i=-1;
+        }
+    }
+}
+
 void Audio::Split(VadModel* vad_obj, vector<std::vector<int>>& vad_segments, bool input_finished)
 {
     AudioFrame *frame;

--
Gitblit v1.9.1