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 "chrome/browser/media/webrtc_browsertest_base.h"
14 #include "chrome/browser/media/webrtc_browsertest_common.h"
15 #include "chrome/browser/profiles/profile.h"
16 #include "chrome/browser/ui/browser.h"
17 #include "chrome/browser/ui/browser_tabstrip.h"
18 #include "chrome/browser/ui/tabs/tab_strip_model.h"
19 #include "chrome/common/chrome_paths.h"
20 #include "chrome/common/chrome_switches.h"
21 #include "chrome/test/base/ui_test_utils.h"
22 #include "content/public/test/browser_test_utils.h"
23 #include "media/base/media_switches.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
{
87 WebRtcAudioQualityBrowserTest() {}
88 virtual void SetUpInProcessBrowserTestFixture() OVERRIDE
{
89 DetectErrorsInJavaScript(); // Look for errors in our rather complex js.
92 virtual void SetUpCommandLine(CommandLine
* command_line
) OVERRIDE
{
93 // This test expects real device handling and requires a real webcam / audio
94 // device; it will not work with fake devices.
95 EXPECT_FALSE(command_line
->HasSwitch(
96 switches::kUseFakeDeviceForMediaStream
));
97 EXPECT_FALSE(command_line
->HasSwitch(
98 switches::kUseFakeUIForMediaStream
));
101 void AddAudioFile(const std::string
& input_file_relative_url
,
102 content::WebContents
* tab_contents
) {
103 EXPECT_EQ("ok-added", ExecuteJavascript(
104 "addAudioFile('" + input_file_relative_url
+ "')", tab_contents
));
107 void PlayAudioFile(content::WebContents
* tab_contents
) {
108 EXPECT_EQ("ok-playing", ExecuteJavascript("playAudioFile()", tab_contents
));
111 base::FilePath
CreateTemporaryWaveFile() {
112 base::FilePath filename
;
113 EXPECT_TRUE(base::CreateTemporaryFile(&filename
));
114 base::FilePath wav_filename
=
115 filename
.AddExtension(FILE_PATH_LITERAL(".wav"));
116 EXPECT_TRUE(base::Move(filename
, wav_filename
));
121 class AudioRecorder
{
123 AudioRecorder(): recording_application_(base::kNullProcessHandle
) {}
126 // Starts the recording program for the specified duration. Returns true
128 bool StartRecording(int duration_sec
, const base::FilePath
& output_file
,
130 EXPECT_EQ(base::kNullProcessHandle
, recording_application_
)
131 << "Tried to record, but is already recording.";
133 CommandLine
command_line(CommandLine::NO_PROGRAM
);
135 // This disable is required to run SoundRecorder.exe on 64-bit Windows
136 // from a 32-bit binary. We need to load the wow64 disable function from
137 // the DLL since it doesn't exist on Windows XP.
138 // TODO(phoglund): find some cleaner solution than using SoundRecorder.exe.
139 base::ScopedNativeLibrary
kernel32_lib(base::FilePath(L
"kernel32"));
140 if (kernel32_lib
.is_valid()) {
141 typedef BOOL (WINAPI
* Wow64DisableWow64FSRedirection
)(PVOID
*);
142 Wow64DisableWow64FSRedirection wow_64_disable_wow_64_fs_redirection
;
143 wow_64_disable_wow_64_fs_redirection
=
144 reinterpret_cast<Wow64DisableWow64FSRedirection
>(
145 kernel32_lib
.GetFunctionPointer(
146 "Wow64DisableWow64FsRedirection"));
147 if (wow_64_disable_wow_64_fs_redirection
!= NULL
) {
148 PVOID
* ignored
= NULL
;
149 wow_64_disable_wow_64_fs_redirection(ignored
);
153 char duration_in_hms
[128] = {0};
154 struct tm duration_tm
= {0};
155 duration_tm
.tm_sec
= duration_sec
;
156 EXPECT_NE(0u, strftime(duration_in_hms
, arraysize(duration_in_hms
),
157 "%H:%M:%S", &duration_tm
));
159 command_line
.SetProgram(
160 base::FilePath(FILE_PATH_LITERAL("SoundRecorder.exe")));
161 command_line
.AppendArg("/FILE");
162 command_line
.AppendArgPath(output_file
);
163 command_line
.AppendArg("/DURATION");
164 command_line
.AppendArg(duration_in_hms
);
166 int num_channels
= mono
? 1 : 2;
167 command_line
.SetProgram(base::FilePath("arecord"));
168 command_line
.AppendArg("-d");
169 command_line
.AppendArg(base::StringPrintf("%d", duration_sec
));
170 command_line
.AppendArg("-f");
171 command_line
.AppendArg("dat");
172 command_line
.AppendArg("-c");
173 command_line
.AppendArg(base::StringPrintf("%d", num_channels
));
174 command_line
.AppendArgPath(output_file
);
177 VLOG(0) << "Running " << command_line
.GetCommandLineString();
178 return base::LaunchProcess(command_line
, base::LaunchOptions(),
179 &recording_application_
);
182 // Joins the recording program. Returns true on success.
183 bool WaitForRecordingToEnd() {
185 base::WaitForExitCode(recording_application_
, &exit_code
);
186 return exit_code
== 0;
189 base::ProcessHandle recording_application_
;
192 bool ForceMicrophoneVolumeTo100Percent() {
194 CommandLine
command_line(test::GetReferenceFilesDir().Append(
195 FILE_PATH_LITERAL("force_mic_volume_max.exe")));
196 VLOG(0) << "Running " << command_line
.GetCommandLineString();
198 if (!base::GetAppOutput(command_line
, &result
)) {
199 LOG(ERROR
) << "Failed to set source volume: output was " << result
;
203 // Just force the volume of, say the first 5 devices. A machine will rarely
204 // have more input sources than that. This is way easier than finding the
205 // input device we happen to be using.
206 for (int device_index
= 0; device_index
< 5; ++device_index
) {
208 const std::string kHundredPercentVolume
= "65536";
209 CommandLine
command_line(base::FilePath(FILE_PATH_LITERAL("pacmd")));
210 command_line
.AppendArg("set-source-volume");
211 command_line
.AppendArg(base::StringPrintf("%d", device_index
));
212 command_line
.AppendArg(kHundredPercentVolume
);
213 VLOG(0) << "Running " << command_line
.GetCommandLineString();
214 if (!base::GetAppOutput(command_line
, &result
)) {
215 LOG(ERROR
) << "Failed to set source volume: output was " << result
;
223 // Removes silence from beginning and end of the |input_audio_file| and writes
224 // the result to the |output_audio_file|. Returns true on success.
225 bool RemoveSilence(const base::FilePath
& input_file
,
226 const base::FilePath
& output_file
) {
227 // SOX documentation for silence command: http://sox.sourceforge.net/sox.html
228 // To remove the silence from both beginning and end of the audio file, we
229 // call sox silence command twice: once on normal file and again on its
230 // reverse, then we reverse the final output.
231 // Silence parameters are (in sequence):
232 // ABOVE_PERIODS: The period for which silence occurs. Value 1 is used for
233 // silence at beginning of audio.
234 // DURATION: the amount of time in seconds that non-silence must be detected
235 // before sox stops trimming audio.
236 // THRESHOLD: value used to indicate what sample value is treates as silence.
237 const char* kAbovePeriods
= "1";
238 const char* kDuration
= "2";
239 const char* kTreshold
= "5%";
242 CommandLine
command_line(test::GetReferenceFilesDir().Append(
243 FILE_PATH_LITERAL("sox.exe")));
245 CommandLine
command_line(base::FilePath(FILE_PATH_LITERAL("sox")));
247 command_line
.AppendArgPath(input_file
);
248 command_line
.AppendArgPath(output_file
);
249 command_line
.AppendArg("silence");
250 command_line
.AppendArg(kAbovePeriods
);
251 command_line
.AppendArg(kDuration
);
252 command_line
.AppendArg(kTreshold
);
253 command_line
.AppendArg("reverse");
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");
260 VLOG(0) << "Running " << command_line
.GetCommandLineString();
262 bool ok
= base::GetAppOutput(command_line
, &result
);
263 VLOG(0) << "Output was:\n\n" << result
;
267 bool CanParseAsFloat(const std::string
& value
) {
268 return atof(value
.c_str()) != 0 || value
== "0";
271 // Runs PESQ to compare |reference_file| to a |actual_file|. The |sample_rate|
272 // can be either 16000 or 8000.
274 // PESQ is only mono-aware, so the files should preferably be recorded in mono.
275 // Furthermore it expects the file to be 16 rather than 32 bits, even though
276 // 32 bits might work. The audio bandwidth of the two files should be the same
277 // e.g. don't compare a 32 kHz file to a 8 kHz file.
279 // The raw score in MOS is written to |raw_mos|, whereas the MOS-LQO score is
280 // written to mos_lqo. The scores are returned as floats in string form (e.g.
281 // "3.145", etc). Returns true on success.
282 bool RunPesq(const base::FilePath
& reference_file
,
283 const base::FilePath
& actual_file
,
284 int sample_rate
, std::string
* raw_mos
, std::string
* mos_lqo
) {
285 // PESQ will break if the paths are too long (!).
286 EXPECT_LT(reference_file
.value().length(), 128u);
287 EXPECT_LT(actual_file
.value().length(), 128u);
290 base::FilePath pesq_path
=
291 test::GetReferenceFilesDir().Append(FILE_PATH_LITERAL("pesq.exe"));
293 base::FilePath pesq_path
=
294 test::GetReferenceFilesDir().Append(FILE_PATH_LITERAL("pesq"));
297 if (!base::PathExists(pesq_path
)) {
298 LOG(ERROR
) << "Missing PESQ binary in " << pesq_path
.value();
302 CommandLine
command_line(pesq_path
);
303 command_line
.AppendArg(base::StringPrintf("+%d", sample_rate
));
304 command_line
.AppendArgPath(reference_file
);
305 command_line
.AppendArgPath(actual_file
);
307 VLOG(0) << "Running " << command_line
.GetCommandLineString();
309 if (!base::GetAppOutput(command_line
, &result
)) {
310 LOG(ERROR
) << "Failed to run PESQ.";
313 VLOG(0) << "Output was:\n\n" << result
;
315 const std::string result_anchor
= "Prediction (Raw MOS, MOS-LQO): = ";
316 std::size_t anchor_pos
= result
.find(result_anchor
);
317 if (anchor_pos
== std::string::npos
) {
318 LOG(ERROR
) << "PESQ was not able to compute a score; we probably recorded "
323 // There are two tab-separated numbers on the format x.xxx, e.g. 5 chars each.
324 std::size_t first_number_pos
= anchor_pos
+ result_anchor
.length();
325 *raw_mos
= result
.substr(first_number_pos
, 5);
326 EXPECT_TRUE(CanParseAsFloat(*raw_mos
)) << "Failed to parse raw MOS number.";
327 *mos_lqo
= result
.substr(first_number_pos
+ 5 + 1, 5);
328 EXPECT_TRUE(CanParseAsFloat(*mos_lqo
)) << "Failed to parse MOS LQO number.";
333 #if defined(OS_LINUX) || defined(OS_WIN)
334 // Only implemented on Linux and Windows for now.
335 #define MAYBE_MANUAL_TestAudioQuality MANUAL_TestAudioQuality
337 #define MAYBE_MANUAL_TestAudioQuality DISABLED_MANUAL_TestAudioQuality
340 IN_PROC_BROWSER_TEST_F(WebRtcAudioQualityBrowserTest
,
341 MAYBE_MANUAL_TestAudioQuality
) {
343 LOG(ERROR
) << "This test is not implemented for Windows XP.";
347 // http://crbug.com/379798.
348 LOG(ERROR
) << "Temporarily disabled for Win 8.";
351 ASSERT_TRUE(test::HasReferenceFilesInCheckout());
352 ASSERT_TRUE(embedded_test_server()->InitializeAndWaitUntilReady());
354 ASSERT_TRUE(ForceMicrophoneVolumeTo100Percent());
356 ui_test_utils::NavigateToURL(
357 browser(), embedded_test_server()->GetURL(kMainWebrtcTestHtmlPage
));
358 content::WebContents
* left_tab
=
359 browser()->tab_strip_model()->GetActiveWebContents();
361 chrome::AddTabAt(browser(), GURL(), -1, true);
362 content::WebContents
* right_tab
=
363 browser()->tab_strip_model()->GetActiveWebContents();
364 ui_test_utils::NavigateToURL(
365 browser(), embedded_test_server()->GetURL(kMainWebrtcTestHtmlPage
));
367 // Prepare the peer connections manually in this test since we don't add
368 // getUserMedia-derived media streams in this test like the other tests.
369 EXPECT_EQ("ok-peerconnection-created",
370 ExecuteJavascript("preparePeerConnection()", left_tab
));
371 EXPECT_EQ("ok-peerconnection-created",
372 ExecuteJavascript("preparePeerConnection()", right_tab
));
374 AddAudioFile(kReferenceFileRelativeUrl
, left_tab
);
376 NegotiateCall(left_tab
, right_tab
);
378 // Note: the media flow isn't necessarily established on the connection just
379 // because the ready state is ok on both sides. We sleep a bit between call
380 // establishment and playing to avoid cutting of the beginning of the audio
382 test::SleepInJavascript(left_tab
, 2000);
384 base::FilePath recording
= CreateTemporaryWaveFile();
386 // Note: the sound clip is about 10 seconds: record for 15 seconds to get some
387 // safety margins on each side.
388 AudioRecorder recorder
;
389 static int kRecordingTimeSeconds
= 15;
390 ASSERT_TRUE(recorder
.StartRecording(kRecordingTimeSeconds
, recording
, true));
392 PlayAudioFile(left_tab
);
394 ASSERT_TRUE(recorder
.WaitForRecordingToEnd());
395 VLOG(0) << "Done recording to " << recording
.value() << std::endl
;
399 base::FilePath trimmed_recording
= CreateTemporaryWaveFile();
401 ASSERT_TRUE(RemoveSilence(recording
, trimmed_recording
));
402 VLOG(0) << "Trimmed silence: " << trimmed_recording
.value() << std::endl
;
406 base::FilePath reference_file_in_test_dir
=
407 test::GetReferenceFilesDir().Append(kReferenceFile
);
408 ASSERT_TRUE(RunPesq(reference_file_in_test_dir
, trimmed_recording
, 16000,
409 &raw_mos
, &mos_lqo
));
411 perf_test::PrintResult("audio_pesq", "", "raw_mos", raw_mos
, "score", true);
412 perf_test::PrintResult("audio_pesq", "", "mos_lqo", mos_lqo
, "score", true);
414 EXPECT_TRUE(base::DeleteFile(recording
, false));
415 EXPECT_TRUE(base::DeleteFile(trimmed_recording
, false));