From bc723ea200144bd6fa8a5dff4b9a780feda144fc Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期四, 29 六月 2023 18:55:01 +0800
Subject: [PATCH] dcos

---
 funasr/runtime/onnxruntime/src/audio.cpp |  191 ++++++++++++++++++++++++++++++++++++-----------
 1 files changed, 145 insertions(+), 46 deletions(-)

diff --git a/funasr/runtime/onnxruntime/src/audio.cpp b/funasr/runtime/onnxruntime/src/audio.cpp
index 635c330..23d0010 100644
--- a/funasr/runtime/onnxruntime/src/audio.cpp
+++ b/funasr/runtime/onnxruntime/src/audio.cpp
@@ -11,6 +11,7 @@
 
 using namespace std;
 
+namespace funasr {
 // see http://soundfile.sapp.org/doc/WaveFormat/
 // Note: We assume little endian here
 struct WaveHeader {
@@ -128,33 +129,32 @@
     start = 0;
 };
 AudioFrame::~AudioFrame(){};
-int AudioFrame::set_start(int val)
+int AudioFrame::SetStart(int val)
 {
     start = val < 0 ? 0 : val;
     return start;
 };
 
-int AudioFrame::set_end(int val)
+int AudioFrame::SetEnd(int val)
 {
     end = val;
     len = end - start;
     return end;
 };
 
-int AudioFrame::get_start()
+int AudioFrame::GetStart()
 {
     return start;
 };
 
-int AudioFrame::get_len()
+int AudioFrame::GetLen()
 {
     return len;
 };
 
-int AudioFrame::disp()
+int AudioFrame::Disp()
 {
-    printf("not imp!!!!\n");
-
+    LOG(ERROR) << "Not imp!!!!";
     return 0;
 };
 
@@ -176,42 +176,37 @@
 {
     if (speech_buff != NULL) {
         free(speech_buff);
-        
     }
-
     if (speech_data != NULL) {
-        
         free(speech_data);
     }
+    if (speech_char != NULL) {
+        free(speech_char);
+    }
 }
 
-void Audio::disp()
+void Audio::Disp()
 {
-    printf("Audio time is %f s. len is %d\n", (float)speech_len / MODEL_SAMPLE_RATE,
-           speech_len);
+    LOG(INFO) << "Audio time is " << (float)speech_len / MODEL_SAMPLE_RATE << " s. len is " << speech_len;
 }
 
-float Audio::get_time_len()
+float Audio::GetTimeLen()
 {
     return (float)speech_len / MODEL_SAMPLE_RATE;
 }
 
-void Audio::wavResample(int32_t sampling_rate, const float *waveform,
+void Audio::WavResample(int32_t sampling_rate, const float *waveform,
                           int32_t n)
 {
-    printf(
-          "Creating a resampler:\n"
-          "   in_sample_rate: %d\n"
-          "   output_sample_rate: %d\n",
-          sampling_rate, static_cast<int32_t>(MODEL_SAMPLE_RATE));
+    LOG(INFO) << "Creating a resampler:\n"
+              << "   in_sample_rate: "<< sampling_rate << "\n"
+              << "   output_sample_rate: " << static_cast<int32_t>(MODEL_SAMPLE_RATE);
     float min_freq =
         std::min<int32_t>(sampling_rate, MODEL_SAMPLE_RATE);
     float lowpass_cutoff = 0.99 * 0.5 * min_freq;
 
     int32_t lowpass_filter_width = 6;
-    //FIXME
-    //auto resampler = new LinearResample(
-    //      sampling_rate, model_sample_rate, lowpass_cutoff, lowpass_filter_width);
+
     auto resampler = std::make_unique<LinearResample>(
           sampling_rate, MODEL_SAMPLE_RATE, lowpass_cutoff, lowpass_filter_width);
     std::vector<float> samples;
@@ -226,7 +221,7 @@
     copy(samples.begin(), samples.end(), speech_data);
 }
 
-bool Audio::loadwav(const char *filename, int32_t* sampling_rate)
+bool Audio::LoadWav(const char *filename, int32_t* sampling_rate)
 {
     WaveHeader header;
     if (speech_data != NULL) {
@@ -240,7 +235,25 @@
     std::ifstream is(filename, std::ifstream::binary);
     is.read(reinterpret_cast<char *>(&header), sizeof(header));
     if(!is){
-        fprintf(stderr, "Failed to read %s\n", filename);
+        LOG(ERROR) << "Failed to read " << filename;
+        return false;
+    }
+
+    if (!header.Validate()) {
+        return false;
+    }
+
+    header.SeekToDataChunk(is);
+    if (!is) {
+        return false;
+    }
+    
+    if (!header.Validate()) {
+        return false;
+    }
+
+    header.SeekToDataChunk(is);
+    if (!is) {
         return false;
     }
     
@@ -255,7 +268,7 @@
         memset(speech_buff, 0, sizeof(int16_t) * speech_len);
         is.read(reinterpret_cast<char *>(speech_buff), header.subchunk2_size);
         if (!is) {
-            fprintf(stderr, "Failed to read %s\n", filename);
+            LOG(ERROR) << "Failed to read " << filename;
             return false;
         }
         speech_data = (float*)malloc(sizeof(float) * speech_len);
@@ -271,7 +284,7 @@
 
         //resample
         if(*sampling_rate != MODEL_SAMPLE_RATE){
-            wavResample(*sampling_rate, speech_data, speech_len);
+            WavResample(*sampling_rate, speech_data, speech_len);
         }
 
         AudioFrame* frame = new AudioFrame(speech_len);
@@ -283,8 +296,47 @@
         return false;
 }
 
-bool Audio::loadwav(const char* buf, int nFileLen, int32_t* sampling_rate)
+bool Audio::LoadWav2Char(const char *filename, int32_t* sampling_rate)
 {
+    WaveHeader header;
+    if (speech_char != NULL) {
+        free(speech_char);
+    }
+    offset = 0;
+    std::ifstream is(filename, std::ifstream::binary);
+    is.read(reinterpret_cast<char *>(&header), sizeof(header));
+    if(!is){
+        LOG(ERROR) << "Failed to read " << filename;
+        return false;
+    }
+    if (!header.Validate()) {
+        return false;
+    }
+    header.SeekToDataChunk(is);
+        if (!is) {
+            return false;
+    }
+    if (!header.Validate()) {
+        return false;
+    }
+    header.SeekToDataChunk(is);
+    if (!is) {
+        return false;
+    }
+    
+    *sampling_rate = header.sample_rate;
+    // header.subchunk2_size contains the number of bytes in the data.
+    // As we assume each sample contains two bytes, so it is divided by 2 here
+    speech_len = header.subchunk2_size / 2;
+    speech_char = (char *)malloc(header.subchunk2_size);
+    memset(speech_char, 0, header.subchunk2_size);
+    is.read(speech_char, header.subchunk2_size);
+
+    return true;
+}
+
+bool Audio::LoadWav(const char* buf, int n_file_len, int32_t* sampling_rate)
+{ 
     WaveHeader header;
     if (speech_data != NULL) {
         free(speech_data);
@@ -318,7 +370,7 @@
         
         //resample
         if(*sampling_rate != MODEL_SAMPLE_RATE){
-            wavResample(*sampling_rate, speech_data, speech_len);
+            WavResample(*sampling_rate, speech_data, speech_len);
         }
 
         AudioFrame* frame = new AudioFrame(speech_len);
@@ -330,7 +382,7 @@
         return false;
 }
 
-bool Audio::loadpcmwav(const char* buf, int nBufLen, int32_t* sampling_rate)
+bool Audio::LoadPcmwav(const char* buf, int n_buf_len, int32_t* sampling_rate)
 {
     if (speech_data != NULL) {
         free(speech_data);
@@ -340,7 +392,7 @@
     }
     offset = 0;
 
-    speech_len = nBufLen / 2;
+    speech_len = n_buf_len / 2;
     speech_buff = (int16_t*)malloc(sizeof(int16_t) * speech_len);
     if (speech_buff)
     {
@@ -361,7 +413,7 @@
         
         //resample
         if(*sampling_rate != MODEL_SAMPLE_RATE){
-            wavResample(*sampling_rate, speech_data, speech_len);
+            WavResample(*sampling_rate, speech_data, speech_len);
         }
 
         AudioFrame* frame = new AudioFrame(speech_len);
@@ -373,7 +425,7 @@
         return false;
 }
 
-bool Audio::loadpcmwav(const char* filename, int32_t* sampling_rate)
+bool Audio::LoadPcmwav(const char* filename, int32_t* sampling_rate)
 {
     if (speech_data != NULL) {
         free(speech_data);
@@ -386,12 +438,15 @@
     FILE* fp;
     fp = fopen(filename, "rb");
     if (fp == nullptr)
+	{
+        LOG(ERROR) << "Failed to read " << filename;
         return false;
+	}
     fseek(fp, 0, SEEK_END);
-    uint32_t nFileLen = ftell(fp);
+    uint32_t n_file_len = ftell(fp);
     fseek(fp, 0, SEEK_SET);
 
-    speech_len = (nFileLen) / 2;
+    speech_len = (n_file_len) / 2;
     speech_buff = (int16_t*)malloc(sizeof(int16_t) * speech_len);
     if (speech_buff)
     {
@@ -412,7 +467,7 @@
 
         //resample
         if(*sampling_rate != MODEL_SAMPLE_RATE){
-            wavResample(*sampling_rate, speech_data, speech_len);
+            WavResample(*sampling_rate, speech_data, speech_len);
         }
 
         AudioFrame* frame = new AudioFrame(speech_len);
@@ -425,7 +480,34 @@
 
 }
 
-int Audio::fetch_chunck(float *&dout, int len)
+bool Audio::LoadPcmwav2Char(const char* filename, int32_t* sampling_rate)
+{
+    if (speech_char != NULL) {
+        free(speech_char);
+    }
+    offset = 0;
+
+    FILE* fp;
+    fp = fopen(filename, "rb");
+    if (fp == nullptr)
+	{
+        LOG(ERROR) << "Failed to read " << filename;
+        return false;
+	}
+    fseek(fp, 0, SEEK_END);
+    uint32_t n_file_len = ftell(fp);
+    fseek(fp, 0, SEEK_SET);
+
+    speech_len = (n_file_len) / 2;
+    speech_char = (char *)malloc(n_file_len);
+    memset(speech_char, 0, n_file_len);
+    fread(speech_char, sizeof(int16_t), n_file_len/2, fp);
+    fclose(fp);
+    
+    return true;
+}
+
+int Audio::FetchChunck(float *&dout, int len)
 {
     if (offset >= speech_align_len) {
         dout = NULL;
@@ -446,14 +528,14 @@
     }
 }
 
-int Audio::fetch(float *&dout, int &len, int &flag)
+int Audio::Fetch(float *&dout, int &len, int &flag)
 {
     if (frame_queue.size() > 0) {
         AudioFrame *frame = frame_queue.front();
         frame_queue.pop();
 
-        dout = speech_data + frame->get_start();
-        len = frame->get_len();
+        dout = speech_data + frame->GetStart();
+        len = frame->GetLen();
         delete frame;
         flag = S_END;
         return 1;
@@ -462,7 +544,7 @@
     }
 }
 
-void Audio::padding()
+void Audio::Padding()
 {
     float num_samples = speech_len;
     float frame_length = 400;
@@ -499,27 +581,44 @@
     delete frame;
 }
 
-void Audio::split(Model* pRecogObj)
+void Audio::Split(OfflineStream* offline_stream)
 {
     AudioFrame *frame;
 
     frame = frame_queue.front();
     frame_queue.pop();
-    int sp_len = frame->get_len();
+    int sp_len = frame->GetLen();
     delete frame;
     frame = NULL;
 
     std::vector<float> pcm_data(speech_data, speech_data+sp_len);
-    vector<std::vector<int>> vad_segments = pRecogObj->vad_seg(pcm_data);
+    vector<std::vector<int>> vad_segments = (offline_stream->vad_handle)->Infer(pcm_data);
     int seg_sample = MODEL_SAMPLE_RATE/1000;
     for(vector<int> segment:vad_segments)
     {
         frame = new AudioFrame();
         int start = segment[0]*seg_sample;
         int end = segment[1]*seg_sample;
-        frame->set_start(start);
-        frame->set_end(end);
+        frame->SetStart(start);
+        frame->SetEnd(end);
         frame_queue.push(frame);
         frame = NULL;
     }
 }
+
+
+void Audio::Split(VadModel* vad_obj, vector<std::vector<int>>& vad_segments, bool input_finished)
+{
+    AudioFrame *frame;
+
+    frame = frame_queue.front();
+    frame_queue.pop();
+    int sp_len = frame->GetLen();
+    delete frame;
+    frame = NULL;
+
+    std::vector<float> pcm_data(speech_data, speech_data+sp_len);
+    vad_segments = vad_obj->Infer(pcm_data, input_finished);
+}
+
+} // namespace funasr
\ No newline at end of file

--
Gitblit v1.9.1