1 // Copyright 2013 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
7 #include "base/command_line.h"
8 #include "base/file_util.h"
9 #include "base/path_service.h"
10 #include "base/process/launch.h"
11 #include "base/scoped_native_library.h"
12 #include "base/strings/stringprintf.h"
13 #include "base/win/windows_version.h"
14 #include "chrome/browser/media/webrtc_browsertest_base.h"
15 #include "chrome/browser/media/webrtc_browsertest_common.h"
16 #include "chrome/browser/profiles/profile.h"
17 #include "chrome/browser/ui/browser.h"
18 #include "chrome/browser/ui/browser_tabstrip.h"
19 #include "chrome/browser/ui/tabs/tab_strip_model.h"
20 #include "chrome/common/chrome_paths.h"
21 #include "chrome/common/chrome_switches.h"
22 #include "chrome/test/base/ui_test_utils.h"
23 #include "content/public/test/browser_test_utils.h"
24 #include "net/test/embedded_test_server/embedded_test_server.h"
25 #include "testing/perf/perf_test.h"
27 // These are relative to the reference file dir defined by
28 // webrtc_browsertest_common.h (i.e. chrome/test/data/webrtc/resources).
29 static const base::FilePath::CharType kReferenceFile[] =
31 FILE_PATH_LITERAL("human-voice-win.wav");
33 FILE_PATH_LITERAL("human-voice-linux.wav");
36 // The javascript will load the reference file relative to its location,
37 // which is in /webrtc on the web server. The files we are looking for are in
38 // webrtc/resources in the chrome/test/data folder.
39 static const char kReferenceFileRelativeUrl[] =
41 "resources/human-voice-win.wav";
43 "resources/human-voice-linux.wav";
46 static const char kMainWebrtcTestHtmlPage[] =
47 "/webrtc/webrtc_audio_quality_test.html";
49 // Test we can set up a WebRTC call and play audio through it.
51 // You must have the src-internal solution in your .gclient to put the required
52 // pyauto_private directory into chrome/test/data/.
54 // This test will only work on machines that have been configured to record
58 // 1. # sudo apt-get install pavucontrol
59 // 2. For the user who will run the test: # pavucontrol
60 // 3. In a separate terminal, # arecord dummy
61 // 4. In pavucontrol, go to the recording tab.
62 // 5. For the ALSA plug-in [aplay]: ALSA Capture from, change from <x> to
63 // <Monitor of x>, where x is whatever your primary sound device is called.
64 // 6. Try launching chrome as the target user on the target machine, try
65 // playing, say, a YouTube video, and record with # arecord -f dat tmp.dat.
66 // Verify the recording with aplay (should have recorded what you played
69 // Note: the volume for ALL your input devices will be forced to 100% by
70 // running this test on Linux.
73 // 1. Control panel > Sound > Manage audio devices.
74 // 2. In the recording tab, right-click in an empty space in the pane with the
75 // devices. Tick 'show disabled devices'.
76 // 3. You should see a 'stero mix' device - this is what your speakers output.
77 // Right click > Properties.
78 // 4. In the Listen tab for the mix device, check the 'listen to this device'
79 // checkbox. Ensure the mix device is the default recording device.
80 // 5. Launch chrome and try playing a video with sound. You should see
81 // in the volume meter for the mix device. Configure the mix device to have
82 // 50 / 100 in level. Also go into the playback tab, right-click Speakers,
83 // and set that level to 50 / 100. Otherwise you will get distortion in
85 class WebRtcAudioQualityBrowserTest : public WebRtcTestBase,
86 public testing::WithParamInterface<bool> {
88 WebRtcAudioQualityBrowserTest() {}
89 virtual void SetUpInProcessBrowserTestFixture() OVERRIDE {
90 DetectErrorsInJavaScript(); // Look for errors in our rather complex js.
93 virtual void SetUpCommandLine(CommandLine* command_line) OVERRIDE {
94 // This test expects real device handling and requires a real webcam / audio
95 // device; it will not work with fake devices.
96 EXPECT_FALSE(command_line->HasSwitch(
97 switches::kUseFakeDeviceForMediaStream));
98 EXPECT_FALSE(command_line->HasSwitch(
99 switches::kUseFakeUIForMediaStream));
101 bool enable_audio_track_processing = GetParam();
102 if (enable_audio_track_processing)
103 command_line->AppendSwitch(switches::kEnableAudioTrackProcessing);
106 void AddAudioFile(const std::string& input_file_relative_url,
107 content::WebContents* tab_contents) {
108 EXPECT_EQ("ok-added", ExecuteJavascript(
109 "addAudioFile('" + input_file_relative_url + "')", tab_contents));
112 void PlayAudioFile(content::WebContents* tab_contents) {
113 EXPECT_EQ("ok-playing", ExecuteJavascript("playAudioFile()", tab_contents));
116 base::FilePath CreateTemporaryWaveFile() {
117 base::FilePath filename;
118 EXPECT_TRUE(base::CreateTemporaryFile(&filename));
119 base::FilePath wav_filename =
120 filename.AddExtension(FILE_PATH_LITERAL(".wav"));
121 EXPECT_TRUE(base::Move(filename, wav_filename));
126 class AudioRecorder {
128 AudioRecorder(): recording_application_(base::kNullProcessHandle) {}
131 // Starts the recording program for the specified duration. Returns true
133 bool StartRecording(int duration_sec, const base::FilePath& output_file,
135 EXPECT_EQ(base::kNullProcessHandle, recording_application_)
136 << "Tried to record, but is already recording.";
138 CommandLine command_line(CommandLine::NO_PROGRAM);
140 // This disable is required to run SoundRecorder.exe on 64-bit Windows
141 // from a 32-bit binary. We need to load the wow64 disable function from
142 // the DLL since it doesn't exist on Windows XP.
143 // TODO(phoglund): find some cleaner solution than using SoundRecorder.exe.
144 base::ScopedNativeLibrary kernel32_lib(base::FilePath(L"kernel32"));
145 if (kernel32_lib.is_valid()) {
146 typedef BOOL (WINAPI* Wow64DisableWow64FSRedirection)(PVOID*);
147 Wow64DisableWow64FSRedirection wow_64_disable_wow_64_fs_redirection;
148 wow_64_disable_wow_64_fs_redirection =
149 reinterpret_cast<Wow64DisableWow64FSRedirection>(
150 kernel32_lib.GetFunctionPointer(
151 "Wow64DisableWow64FsRedirection"));
152 if (wow_64_disable_wow_64_fs_redirection != NULL) {
153 PVOID* ignored = NULL;
154 wow_64_disable_wow_64_fs_redirection(ignored);
158 char duration_in_hms[128] = {0};
159 struct tm duration_tm = {0};
160 duration_tm.tm_sec = duration_sec;
161 EXPECT_NE(0u, strftime(duration_in_hms, arraysize(duration_in_hms),
162 "%H:%M:%S", &duration_tm));
164 command_line.SetProgram(
165 base::FilePath(FILE_PATH_LITERAL("SoundRecorder.exe")));
166 command_line.AppendArg("/FILE");
167 command_line.AppendArgPath(output_file);
168 command_line.AppendArg("/DURATION");
169 command_line.AppendArg(duration_in_hms);
171 int num_channels = mono ? 1 : 2;
172 command_line.SetProgram(base::FilePath("arecord"));
173 command_line.AppendArg("-d");
174 command_line.AppendArg(base::StringPrintf("%d", duration_sec));
175 command_line.AppendArg("-f");
176 command_line.AppendArg("dat");
177 command_line.AppendArg("-c");
178 command_line.AppendArg(base::StringPrintf("%d", num_channels));
179 command_line.AppendArgPath(output_file);
182 VLOG(0) << "Running " << command_line.GetCommandLineString();
183 return base::LaunchProcess(command_line, base::LaunchOptions(),
184 &recording_application_);
187 // Joins the recording program. Returns true on success.
188 bool WaitForRecordingToEnd() {
190 base::WaitForExitCode(recording_application_, &exit_code);
191 return exit_code == 0;
194 base::ProcessHandle recording_application_;
197 bool ForceMicrophoneVolumeTo100Percent() {
199 CommandLine command_line(test::GetReferenceFilesDir().Append(
200 FILE_PATH_LITERAL("force_mic_volume_max.exe")));
201 VLOG(0) << "Running " << command_line.GetCommandLineString();
203 if (!base::GetAppOutput(command_line, &result)) {
204 LOG(ERROR) << "Failed to set source volume: output was " << result;
208 // Just force the volume of, say the first 5 devices. A machine will rarely
209 // have more input sources than that. This is way easier than finding the
210 // input device we happen to be using.
211 for (int device_index = 0; device_index < 5; ++device_index) {
213 const std::string kHundredPercentVolume = "65536";
214 CommandLine command_line(base::FilePath(FILE_PATH_LITERAL("pacmd")));
215 command_line.AppendArg("set-source-volume");
216 command_line.AppendArg(base::StringPrintf("%d", device_index));
217 command_line.AppendArg(kHundredPercentVolume);
218 VLOG(0) << "Running " << command_line.GetCommandLineString();
219 if (!base::GetAppOutput(command_line, &result)) {
220 LOG(ERROR) << "Failed to set source volume: output was " << result;
228 // Removes silence from beginning and end of the |input_audio_file| and writes
229 // the result to the |output_audio_file|. Returns true on success.
230 bool RemoveSilence(const base::FilePath& input_file,
231 const base::FilePath& output_file) {
232 // SOX documentation for silence command: http://sox.sourceforge.net/sox.html
233 // To remove the silence from both beginning and end of the audio file, we
234 // call sox silence command twice: once on normal file and again on its
235 // reverse, then we reverse the final output.
236 // Silence parameters are (in sequence):
237 // ABOVE_PERIODS: The period for which silence occurs. Value 1 is used for
238 // silence at beginning of audio.
239 // DURATION: the amount of time in seconds that non-silence must be detected
240 // before sox stops trimming audio.
241 // THRESHOLD: value used to indicate what sample value is treates as silence.
242 const char* kAbovePeriods = "1";
243 const char* kDuration = "2";
244 const char* kTreshold = "5%";
247 CommandLine command_line(test::GetReferenceFilesDir().Append(
248 FILE_PATH_LITERAL("sox.exe")));
250 CommandLine command_line(base::FilePath(FILE_PATH_LITERAL("sox")));
252 command_line.AppendArgPath(input_file);
253 command_line.AppendArgPath(output_file);
254 command_line.AppendArg("silence");
255 command_line.AppendArg(kAbovePeriods);
256 command_line.AppendArg(kDuration);
257 command_line.AppendArg(kTreshold);
258 command_line.AppendArg("reverse");
259 command_line.AppendArg("silence");
260 command_line.AppendArg(kAbovePeriods);
261 command_line.AppendArg(kDuration);
262 command_line.AppendArg(kTreshold);
263 command_line.AppendArg("reverse");
265 VLOG(0) << "Running " << command_line.GetCommandLineString();
267 bool ok = base::GetAppOutput(command_line, &result);
268 VLOG(0) << "Output was:\n\n" << result;
272 bool CanParseAsFloat(const std::string& value) {
273 return atof(value.c_str()) != 0 || value == "0";
276 // Runs PESQ to compare |reference_file| to a |actual_file|. The |sample_rate|
277 // can be either 16000 or 8000.
279 // PESQ is only mono-aware, so the files should preferably be recorded in mono.
280 // Furthermore it expects the file to be 16 rather than 32 bits, even though
281 // 32 bits might work. The audio bandwidth of the two files should be the same
282 // e.g. don't compare a 32 kHz file to a 8 kHz file.
284 // The raw score in MOS is written to |raw_mos|, whereas the MOS-LQO score is
285 // written to mos_lqo. The scores are returned as floats in string form (e.g.
286 // "3.145", etc). Returns true on success.
287 bool RunPesq(const base::FilePath& reference_file,
288 const base::FilePath& actual_file,
289 int sample_rate, std::string* raw_mos, std::string* mos_lqo) {
290 // PESQ will break if the paths are too long (!).
291 EXPECT_LT(reference_file.value().length(), 128u);
292 EXPECT_LT(actual_file.value().length(), 128u);
295 base::FilePath pesq_path =
296 test::GetReferenceFilesDir().Append(FILE_PATH_LITERAL("pesq.exe"));
298 base::FilePath pesq_path =
299 test::GetReferenceFilesDir().Append(FILE_PATH_LITERAL("pesq"));
302 if (!base::PathExists(pesq_path)) {
303 LOG(ERROR) << "Missing PESQ binary in " << pesq_path.value();
307 CommandLine command_line(pesq_path);
308 command_line.AppendArg(base::StringPrintf("+%d", sample_rate));
309 command_line.AppendArgPath(reference_file);
310 command_line.AppendArgPath(actual_file);
312 VLOG(0) << "Running " << command_line.GetCommandLineString();
314 if (!base::GetAppOutput(command_line, &result)) {
315 LOG(ERROR) << "Failed to run PESQ.";
318 VLOG(0) << "Output was:\n\n" << result;
320 const std::string result_anchor = "Prediction (Raw MOS, MOS-LQO): = ";
321 std::size_t anchor_pos = result.find(result_anchor);
322 if (anchor_pos == std::string::npos) {
323 LOG(ERROR) << "PESQ was not able to compute a score; we probably recorded "
328 // There are two tab-separated numbers on the format x.xxx, e.g. 5 chars each.
329 std::size_t first_number_pos = anchor_pos + result_anchor.length();
330 *raw_mos = result.substr(first_number_pos, 5);
331 EXPECT_TRUE(CanParseAsFloat(*raw_mos)) << "Failed to parse raw MOS number.";
332 *mos_lqo = result.substr(first_number_pos + 5 + 1, 5);
333 EXPECT_TRUE(CanParseAsFloat(*mos_lqo)) << "Failed to parse MOS LQO number.";
338 static const bool kRunTestsWithFlag[] = { false, true };
339 INSTANTIATE_TEST_CASE_P(WebRtcAudioQualityBrowserTests,
340 WebRtcAudioQualityBrowserTest,
341 testing::ValuesIn(kRunTestsWithFlag));
343 #if defined(OS_LINUX) || defined(OS_WIN)
344 // Only implemented on Linux and Windows for now.
345 #define MAYBE_MANUAL_TestAudioQuality MANUAL_TestAudioQuality
347 #define MAYBE_MANUAL_TestAudioQuality DISABLED_MANUAL_TestAudioQuality
350 IN_PROC_BROWSER_TEST_P(WebRtcAudioQualityBrowserTest,
351 MAYBE_MANUAL_TestAudioQuality) {
353 if (base::win::GetVersion() < base::win::VERSION_VISTA) {
354 // It would take work to implement this on XP; not prioritized right now.
355 LOG(ERROR) << "This test is not implemented for Windows XP.";
359 ASSERT_TRUE(test::HasReferenceFilesInCheckout());
360 ASSERT_TRUE(embedded_test_server()->InitializeAndWaitUntilReady());
362 ASSERT_TRUE(ForceMicrophoneVolumeTo100Percent());
364 ui_test_utils::NavigateToURL(
365 browser(), embedded_test_server()->GetURL(kMainWebrtcTestHtmlPage));
366 content::WebContents* left_tab =
367 browser()->tab_strip_model()->GetActiveWebContents();
369 chrome::AddTabAt(browser(), GURL(), -1, true);
370 content::WebContents* right_tab =
371 browser()->tab_strip_model()->GetActiveWebContents();
372 ui_test_utils::NavigateToURL(
373 browser(), embedded_test_server()->GetURL(kMainWebrtcTestHtmlPage));
375 // Prepare the peer connections manually in this test since we don't add
376 // getUserMedia-derived media streams in this test like the other tests.
377 EXPECT_EQ("ok-peerconnection-created",
378 ExecuteJavascript("preparePeerConnection()", left_tab));
379 EXPECT_EQ("ok-peerconnection-created",
380 ExecuteJavascript("preparePeerConnection()", right_tab));
382 AddAudioFile(kReferenceFileRelativeUrl, left_tab);
384 NegotiateCall(left_tab, right_tab);
386 // Note: the media flow isn't necessarily established on the connection just
387 // because the ready state is ok on both sides. We sleep a bit between call
388 // establishment and playing to avoid cutting of the beginning of the audio
390 test::SleepInJavascript(left_tab, 2000);
392 base::FilePath recording = CreateTemporaryWaveFile();
394 // Note: the sound clip is about 10 seconds: record for 15 seconds to get some
395 // safety margins on each side.
396 AudioRecorder recorder;
397 static int kRecordingTimeSeconds = 15;
398 ASSERT_TRUE(recorder.StartRecording(kRecordingTimeSeconds, recording, true));
400 PlayAudioFile(left_tab);
402 ASSERT_TRUE(recorder.WaitForRecordingToEnd());
403 VLOG(0) << "Done recording to " << recording.value() << std::endl;
407 base::FilePath trimmed_recording = CreateTemporaryWaveFile();
409 ASSERT_TRUE(RemoveSilence(recording, trimmed_recording));
410 VLOG(0) << "Trimmed silence: " << trimmed_recording.value() << std::endl;
414 base::FilePath reference_file_in_test_dir =
415 test::GetReferenceFilesDir().Append(kReferenceFile);
416 ASSERT_TRUE(RunPesq(reference_file_in_test_dir, trimmed_recording, 16000,
417 &raw_mos, &mos_lqo));
419 perf_test::PrintResult("audio_pesq", "", "raw_mos", raw_mos, "score", true);
420 perf_test::PrintResult("audio_pesq", "", "mos_lqo", mos_lqo, "score", true);
422 EXPECT_TRUE(base::DeleteFile(recording, false));
423 EXPECT_TRUE(base::DeleteFile(trimmed_recording, false));