Allow intelligibility to compile in apm

webrtc/modules/audio_processing/intelligibility/intelligibility_proc.cc

 /*
  *  Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#include <arpa/inet.h>
-#include <fcntl.h>
+//
+//  Command line tool for speech intelligibility enhancement. Provides for
+//  running and testing intelligibility_enhancer as an independent process.
+//  Use --help for options.
+//
+
 #include <stdint.h>
-#include <stdio.h>
 #include <stdlib.h>
-#include <sys/mman.h>
+#include <string>
 #include <sys/stat.h>
 #include <sys/types.h>
-#include <unistd.h>
-
-#include <fenv.h>
-#include <limits>
-
-#include <complex>
 
 #include "gflags/gflags.h"
+#include "testing/gtest/include/gtest/gtest.h"
 #include "webrtc/base/checks.h"
 #include "webrtc/common_audio/real_fourier.h"
+#include "webrtc/common_audio/wav_file.h"
 #include "webrtc/modules/audio_processing/intelligibility/intelligibility_enhancer.h"
 #include "webrtc/modules/audio_processing/intelligibility/intelligibility_utils.h"
 #include "webrtc/system_wrappers/interface/critical_section_wrapper.h"
-#include "webrtc/system_wrappers/interface/scoped_ptr.h"
 
+// PCM data simulating streams
 const int16_t* in_ipcm;

[Andrew MacDonald, 2015/06/18 03:38:15]

I don't think any of these from here down to line 43 need to be globals, right?

[ekm, 2015/06/18 18:50:27]

Done.

-int16_t* out_ipcm;
-const int16_t* noise_ipcm;
-
 float* in_fpcm;
 float* out_fpcm;
 float* noise_fpcm;
+
+// Current locations in streams
 float* noise_cursor;
 float* clear_cursor;
 
-int samples;
-int fragment_size;
+size_t samples;        // Number of samples in input PCM file
+size_t fragment_size;  // Number of samples to process at a time
+                    // to simulate APM stream processing
 
 using std::complex;
+
+namespace webrtc {
+
 using webrtc::RealFourier;
 using webrtc::IntelligibilityEnhancer;
 
-DEFINE_int32(clear_type, webrtc::intelligibility::VarianceArray::kStepInfinite,
+DEFINE_int32(clear_type,
+             webrtc::intelligibility::VarianceArray::kStepInfinite,
              "Variance algorithm for clear data.");
-DEFINE_double(clear_alpha, 0.9,
-              "Variance decay factor for clear data.");
-DEFINE_int32(clear_window, 475,
+DEFINE_double(clear_alpha, 0.9, "Variance decay factor for clear data.");
+DEFINE_int32(clear_window,
+             475,
              "Window size for windowed variance for clear data.");
-DEFINE_int32(sample_rate, 16000,
+DEFINE_int32(sample_rate,
+             16000,
              "Audio sample rate used in the input and output files.");
-DEFINE_int32(ana_rate, 800,
+DEFINE_int32(ana_rate,
+             800,
              "Analysis rate; gains recalculated every N blocks.");
-DEFINE_int32(var_rate, 2,
-             "Variance clear rate; history is forgotten every N gain recalculations.");
+DEFINE_int32(
+    var_rate,
+    2,
+    "Variance clear rate; history is forgotten every N gain recalculations.");
 DEFINE_double(gain_limit, 1000.0, "Maximum gain change in one block.");
 
 DEFINE_bool(repeat, false, "Repeat input file ad nauseam.");
 
-DEFINE_string(clear_file, "speech.pcm", "Input file with clear speech.");
-DEFINE_string(noise_file, "noise.pcm", "Input file with noise data.");
-DEFINE_string(out_file, "proc_enhanced.pcm", "Enhanced output. Use '-' to "
-              "pipe through aplay internally.");
+DEFINE_string(clear_file, "speech.wav", "Input file with clear speech.");
+DEFINE_string(noise_file, "noise.wav", "Input file with noise data.");
+DEFINE_string(out_file,
+              "proc_enhanced.wav",
+              "Enhanced output. Use '-' to "
+              "play through aplay immediately.");
 
-// Write an Sun AU-formatted audio chunk into file descriptor |fd|. Can be used
-// to pipe the audio stream directly into aplay.
-void writeau(int fd) {
-  uint32_t thing;
+// Constant IntelligibilityEnhancer constructor parameters.
+const int kErbResolution = 2;
+const int kNumChannels = 1;
 
-  write(fd, ".snd", 4);
-  thing = htonl(24);
-  write(fd, &thing, sizeof(thing));
-  thing = htonl(0xffffffff);
-  write(fd, &thing, sizeof(thing));
-  thing = htonl(3);
-  write(fd, &thing, sizeof(thing));
-  thing = htonl(FLAGS_sample_rate);
-  write(fd, &thing, sizeof(thing));
-  thing = htonl(1);
-  write(fd, &thing, sizeof(thing));
+const std::string kTmpOutName = "tmp_out_file.wav";

[Andrew MacDonald, 2015/06/18 03:38:15]

Prefer not to use non-POD types as static globals. Either make this a const
char* or move it into void_main.

[ekm, 2015/06/18 18:50:26]

Done.

 
-  for (int i = 0; i < samples; ++i) {
-    out_ipcm[i] = htons(out_ipcm[i]);
-  }
-  write(fd, out_ipcm, sizeof(*out_ipcm) * samples);
-}
-
-int main(int argc, char* argv[]) {
-  google::SetUsageMessage("\n\nVariance algorithm types are:\n"
-                          "  0 - infinite/normal,\n"
-                          "  1 - exponentially decaying,\n"
-                          "  2 - rolling window.\n"
-                          "\nInput files must be little-endian 16-bit signed raw PCM.\n");
+// void function for gtest
+void void_main(int argc, char* argv[]) {
+  google::SetUsageMessage(
+      "\n\nVariance algorithm types are:\n"
+      "  0 - infinite/normal,\n"
+      "  1 - exponentially decaying,\n"
+      "  2 - rolling window.\n"
+      "\nInput files must be little-endian 16-bit signed raw PCM.\n");
   google::ParseCommandLineFlags(&argc, &argv, true);
 
   const char* in_name = FLAGS_clear_file.c_str();
   const char* out_name = FLAGS_out_file.c_str();
   const char* noise_name = FLAGS_noise_file.c_str();
   struct stat in_stat, noise_stat;
-  int in_fd, out_fd, noise_fd;
-  FILE* aplay_file = nullptr;
+  std::string out_file_name;
+  std::string in_file_name = in_name;
+  std::string noise_file_name = noise_name;

[Andrew MacDonald, 2015/06/18 03:38:15]

You have three variables referring to each of the filenames. I think you can get
away with just using the FLAGS variant everywhere.

[ekm, 2015/06/18 18:50:26]

Done.

 
-  fragment_size = FLAGS_sample_rate / 100;
+  // Load settings and wav input.
 
-  stat(in_name, &in_stat);
-  stat(noise_name, &noise_stat);
+  fragment_size = FLAGS_sample_rate / 100;  // Mirror real time APM chunk size.
+                                            // Duplicates chunk_length_ in
+                                            // IntelligibilityEnhancer.
+
+  ASSERT_EQ(stat(in_name, &in_stat), 0) << "Empty speech input.";

[Andrew MacDonald, 2015/06/18 03:38:15]

Not sure these are really needed. How often will you have any empty file?

[ekm, 2015/06/18 18:50:27]

Done.

+  ASSERT_EQ(stat(noise_name, &noise_stat), 0) << "Empty noise input.";
+
   samples = in_stat.st_size / sizeof(*in_ipcm);
 
-  in_fd = open(in_name, O_RDONLY);
   if (!strcmp(out_name, "-")) {
-    aplay_file = popen("aplay -t au", "w");
-    out_fd = fileno(aplay_file);
+    out_file_name = kTmpOutName;
   } else {
-    out_fd = open(out_name, O_WRONLY | O_CREAT | O_TRUNC,
-                  S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
-  }
-  noise_fd = open(noise_name, O_RDONLY);
-
-  in_ipcm = static_cast<int16_t*>(mmap(nullptr, in_stat.st_size, PROT_READ,
-                                       MAP_PRIVATE, in_fd, 0));
-  noise_ipcm = static_cast<int16_t*>(mmap(nullptr, noise_stat.st_size,
-                                          PROT_READ, MAP_PRIVATE, noise_fd, 0));
-  out_ipcm = new int16_t[samples];
-  out_fpcm = new float[samples];
-  in_fpcm = new float[samples];
-  noise_fpcm = new float[samples];
-
-  for (int i = 0; i < samples; ++i) {
-    noise_fpcm[i] = noise_ipcm[i % (noise_stat.st_size / sizeof(*noise_ipcm))];
+    out_file_name = out_name;
   }
 
-  //feenableexcept(FE_INVALID | FE_OVERFLOW);
-  IntelligibilityEnhancer enh(2,
-                              FLAGS_sample_rate, 1,
-                              FLAGS_clear_type,
-                              static_cast<float>(FLAGS_clear_alpha),
-                              FLAGS_clear_window,
-                              FLAGS_ana_rate,
-                              FLAGS_var_rate,
-                              FLAGS_gain_limit);
+  WavReader in_file(in_file_name);
+  in_fpcm = new float[samples];

[Andrew MacDonald, 2015/06/18 03:38:15]

vector or at least scoped_ptr. Doesn't matter if we leak here of course, but
let's practice good habits ;)

[ekm, 2015/06/18 18:50:26]

Done.

+  const size_t samples_read = in_file.ReadSamples(samples, in_fpcm);
+
+  WavReader noise_file(noise_file_name);
+  noise_fpcm = new float[samples];

[Andrew MacDonald, 2015/06/18 03:38:15]

vector

[ekm, 2015/06/18 18:50:27]

Done.

+  const size_t noise_samples = noise_file.ReadSamples(samples_read, noise_fpcm);
+  for (size_t i = noise_samples; i < samples_read; ++i) {
+    noise_fpcm[i] = noise_fpcm[i % noise_samples];

[Andrew MacDonald, 2015/06/18 03:38:15]

This is a bit weird. You're restarting the noise file, which may not be a great
idea, depending on the content.

I think better to crop to the shorter of the two files.

[ekm, 2015/06/18 18:50:27]

For iid noise it's fine, but I see with real world noise things could get weird.
Done.

+  }
+
+  // Run intelligibility enhancement.
+
+  IntelligibilityEnhancer enh(
+      kErbResolution,
+      FLAGS_sample_rate,
+      kNumChannels,
+      FLAGS_clear_type, static_cast<float>(FLAGS_clear_alpha),
+      FLAGS_clear_window, FLAGS_ana_rate, FLAGS_var_rate, FLAGS_gain_limit);
 
   // Slice the input into smaller chunks, as the APM would do, and feed them
-  // into the enhancer. Repeat indefinitely if FLAGS_repeat is set.
+  // through the enhancer. Repeat indefinitely if FLAGS_repeat is set.
   do {
+    clear_cursor = in_fpcm;
     noise_cursor = noise_fpcm;
-    clear_cursor = in_fpcm;
-    for (int i = 0; i < samples; ++i) {
-      in_fpcm[i] = in_ipcm[i];
-    }
 
-    for (int i = 0; i < samples; i += fragment_size) {
+    for (size_t i = 0; i < samples; i += fragment_size) {

[Andrew MacDonald, 2015/06/18 03:38:15]

I think it's more typical to read directly from the file in chunks, but this is
OK.

[ekm, 2015/06/18 18:50:26]

Acknowledged.

       enh.ProcessCaptureAudio(&noise_cursor);
       enh.ProcessRenderAudio(&clear_cursor);
       clear_cursor += fragment_size;
       noise_cursor += fragment_size;
     }
 
-    for (int i = 0; i < samples; ++i) {
-      out_ipcm[i] = static_cast<float>(in_fpcm[i]);
+    {
+    WavWriter out_file(out_file_name, FLAGS_sample_rate, kNumChannels);

[Andrew MacDonald, 2015/06/18 03:38:15]

Indent inside scopes.

[ekm, 2015/06/18 18:50:26]

Done.

+    out_file.WriteSamples(in_fpcm, samples);
     }
+
     if (!strcmp(out_name, "-")) {
-      writeau(out_fd);
-    } else {
-      write(out_fd, out_ipcm, samples * sizeof(*out_ipcm));
+      system(("aplay " + out_file_name).c_str());
+      system(("rm " + out_file_name).c_str());
     }
+
   } while (FLAGS_repeat);

[Andrew MacDonald, 2015/06/18 03:38:15]

This probably made sense with the piping directly to aplay. But now that you
write the whole thing to file then read, you're doing to have a gap between the
repeats. I think better to just remove this option.

Not sure this is worth it, but if you want the same behavior, you could open a
fifo:
http://linux.die.net/man/3/mkfifo

Write to it using WavWriter and read from it using aplay.

[ekm, 2015/06/18 18:50:26]

Done. Removed. I don't think it's worth it.

 
-  munmap(const_cast<int16_t*>(noise_ipcm), noise_stat.st_size);
-  munmap(const_cast<int16_t*>(in_ipcm), in_stat.st_size);
-  close(noise_fd);
-  if (aplay_file) {
-    pclose(aplay_file);
-  } else {
-    close(out_fd);
-  }
-  close(in_fd);
+}
 
+}  // namespace webrtc
+
+int main(int argc, char* argv[]) {
+  webrtc::void_main(argc, argv);
   return 0;
 }
-