From a0048dc766f233827f2e7b5ebed0a0e22fae44b1 Mon Sep 17 00:00:00 2001
From: Yabin Li <wucong.lyb@alibaba-inc.com>
Date: 星期三, 13 十二月 2023 09:50:58 +0800
Subject: [PATCH] 8k (#1174) (#1175)

---
 runtime/onnxruntime/src/audio.cpp |   44 +++++++++++++++++++++++++++-----------------
 1 files changed, 27 insertions(+), 17 deletions(-)

diff --git a/runtime/onnxruntime/src/audio.cpp b/runtime/onnxruntime/src/audio.cpp
index b543797..132f47d 100644
--- a/runtime/onnxruntime/src/audio.cpp
+++ b/runtime/onnxruntime/src/audio.cpp
@@ -193,18 +193,28 @@
     return 0;
 }
 
-Audio::Audio(int data_type) : data_type(data_type)
+Audio::Audio(int data_type) : dest_sample_rate(MODEL_SAMPLE_RATE), data_type(data_type)
 {
     speech_buff = NULL;
     speech_data = NULL;
     align_size = 1360;
+    seg_sample = dest_sample_rate / 1000;
 }
 
-Audio::Audio(int data_type, int size) : data_type(data_type)
+Audio::Audio(int model_sample_rate, int data_type) : dest_sample_rate(model_sample_rate), data_type(data_type)
+{
+    speech_buff = NULL;
+    speech_data = NULL;
+    align_size = 1360;
+    seg_sample = dest_sample_rate / 1000;
+}
+
+Audio::Audio(int model_sample_rate, int data_type, int size) : dest_sample_rate(model_sample_rate), data_type(data_type)
 {
     speech_buff = NULL;
     speech_data = NULL;
     align_size = (float)size;
+    seg_sample = dest_sample_rate / 1000;
 }
 
 Audio::~Audio()
@@ -222,12 +232,12 @@
 
 void Audio::Disp()
 {
-    LOG(INFO) << "Audio time is " << (float)speech_len / MODEL_SAMPLE_RATE << " s. len is " << speech_len;
+    LOG(INFO) << "Audio time is " << (float)speech_len / dest_sample_rate << " s. len is " << speech_len;
 }
 
 float Audio::GetTimeLen()
 {
-    return (float)speech_len / MODEL_SAMPLE_RATE;
+    return (float)speech_len / dest_sample_rate;
 }
 
 void Audio::WavResample(int32_t sampling_rate, const float *waveform,
@@ -237,13 +247,13 @@
               << "   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);
+        std::min<int32_t>(sampling_rate, dest_sample_rate);
     float lowpass_cutoff = 0.99 * 0.5 * min_freq;
 
     int32_t lowpass_filter_width = 6;
 
     auto resampler = std::make_unique<LinearResample>(
-          sampling_rate, MODEL_SAMPLE_RATE, lowpass_cutoff, lowpass_filter_width);
+          sampling_rate, dest_sample_rate, lowpass_cutoff, lowpass_filter_width);
     std::vector<float> samples;
     resampler->Resample(waveform, n, true, &samples);
     //reset speech_data
@@ -311,7 +321,7 @@
         nullptr, // allocate a new context
         AV_CH_LAYOUT_MONO, // output channel layout (stereo)
         AV_SAMPLE_FMT_S16, // output sample format (signed 16-bit)
-        16000, // output sample rate (same as input)
+        dest_sample_rate, // output sample rate (same as input)
         av_get_default_channel_layout(codecContext->channels), // input channel layout
         codecContext->sample_fmt, // input sample format
         codecContext->sample_rate, // input sample rate
@@ -347,7 +357,7 @@
                     int in_samples = frame->nb_samples;
                     uint8_t **in_data = frame->extended_data;
                     int out_samples = av_rescale_rnd(in_samples,
-                                                    16000,
+                                                    dest_sample_rate,
                                                     codecContext->sample_rate,
                                                     AV_ROUND_DOWN);
                     
@@ -494,7 +504,7 @@
         nullptr, // allocate a new context
         AV_CH_LAYOUT_MONO, // output channel layout (stereo)
         AV_SAMPLE_FMT_S16, // output sample format (signed 16-bit)
-        16000, // output sample rate (same as input)
+        dest_sample_rate, // output sample rate (same as input)
         av_get_default_channel_layout(codecContext->channels), // input channel layout
         codecContext->sample_fmt, // input sample format
         codecContext->sample_rate, // input sample rate
@@ -532,7 +542,7 @@
                     int in_samples = frame->nb_samples;
                     uint8_t **in_data = frame->extended_data;
                     int out_samples = av_rescale_rnd(in_samples,
-                                                    16000,
+                                                    dest_sample_rate,
                                                     codecContext->sample_rate,
                                                     AV_ROUND_DOWN);
                     
@@ -666,7 +676,7 @@
         }
 
         //resample
-        if(*sampling_rate != MODEL_SAMPLE_RATE){
+        if(*sampling_rate != dest_sample_rate){
             WavResample(*sampling_rate, speech_data, speech_len);
         }
 
@@ -752,7 +762,7 @@
         }
         
         //resample
-        if(*sampling_rate != MODEL_SAMPLE_RATE){
+        if(*sampling_rate != dest_sample_rate){
             WavResample(*sampling_rate, speech_data, speech_len);
         }
 
@@ -795,7 +805,7 @@
         }
         
         //resample
-        if(*sampling_rate != MODEL_SAMPLE_RATE){
+        if(*sampling_rate != dest_sample_rate){
             WavResample(*sampling_rate, speech_data, speech_len);
         }
 
@@ -840,7 +850,7 @@
         }
         
         //resample
-        if(*sampling_rate != MODEL_SAMPLE_RATE){
+        if(*sampling_rate != dest_sample_rate){
             WavResample(*sampling_rate, speech_data, speech_len);
         }
 
@@ -898,7 +908,7 @@
         }
 
         //resample
-        if(*sampling_rate != MODEL_SAMPLE_RATE){
+        if(*sampling_rate != dest_sample_rate){
             WavResample(*sampling_rate, speech_data, speech_len);
         }
 
@@ -1009,7 +1019,7 @@
         AudioFrame *frame = frame_queue.front();
         frame_queue.pop();
 
-        start_time = (float)(frame->GetStart())/MODEL_SAMPLE_RATE;
+        start_time = (float)(frame->GetStart())/ dest_sample_rate;
         dout = speech_data + frame->GetStart();
         len = frame->GetLen();
         delete frame;
@@ -1248,7 +1258,7 @@
     }
 
     // erase all_samples
-    int vector_cache = MODEL_SAMPLE_RATE*2;
+    int vector_cache = dest_sample_rate*2;
     if(speech_offline_start == -1){
         if(all_samples.size() > vector_cache){
             int erase_num = all_samples.size() - vector_cache;

--
Gitblit v1.9.1