# Lint as: python2, python3 # Copyright 2021 The Chromium OS Authors. All rights reserved. # Use of this source code is governed by a BSD-style license that can be # found in the LICENSE file. """This is a server side noise cancellation test using the Chameleon board.""" import logging import os import time from autotest_lib.client.common_lib import error from autotest_lib.client.cros.audio import audio_test_data from autotest_lib.client.cros.audio import sox_utils from autotest_lib.client.cros.audio import visqol_utils from autotest_lib.client.cros.bluetooth.bluetooth_audio_test_data import ( download_file_from_bucket, get_visqol_binary) from autotest_lib.client.cros.chameleon import audio_test_utils from autotest_lib.client.cros.chameleon import chameleon_audio_ids from autotest_lib.client.cros.chameleon import chameleon_audio_helper from autotest_lib.server.cros.audio import audio_test DIST_FILES_DIR = 'gs://chromeos-localmirror/distfiles/test_noise_cancellation' DATA_DIR = '/tmp' # Verification steps for the Noise Cancellation processing (NC): # 1. Prepare the audio source file and reference file. # 2. Play the source file by Chameleon. # 3. Record by DUT Internal Mic when NC is on and get ViSQOL score A. # 4. Repeat step 2. # 5. Record by DUT Internal Mic when NC is off and get ViSQOL score B. # 6. Check if A - B >= threshold # # In practice, ViSQOL is not the most suitable metrics for NC due to its # intrusive design (reference: go/visqol). However, it is fair enough to compare # the relative gain (or degradation) between before and after de-noising. # # TODO(johnylin): replace ViSQOL with other metrics if applicable. # TODO(johnylin): add more speech and noise test inputs for inclusion. class audio_AudioNoiseCancellation(audio_test.AudioTest): """Server side input audio noise cancellation test. This test talks to a Chameleon board and a Cros device to verify input audio noise cancellation function of the Cros device. """ version = 1 DELAY_BEFORE_PLAYBACK_SECONDS = 3.0 DELAY_AFTER_PLAYBACK_SECONDS = 2.0 DELAY_AFTER_BINDING = 0.5 DELAY_AFTER_NC_TOGGLED = 0.5 cleanup_files = [] def cleanup(self): # Restore the default state of bypass blocking mechanism in Cras. # Restarting Cras is only way because we are not able to know the # default state. self.host.run('restart cras') # Start Chrome UI. self.host.run('start ui') # Remove downloaded files and the temporary generated files. for cleanup_file in self.cleanup_files: if os.path.isfile(cleanup_file): os.remove(cleanup_file) def download_file_from_bucket(self, file): """Download the file from GS bucket. @param file: the file name for download. @raises: error.TestError if failed. @returns: the local path of the downloaded file. """ remote_path = os.path.join(DIST_FILES_DIR, file) if not download_file_from_bucket( DATA_DIR, remote_path, lambda _, __, p: p.returncode == 0): logging.error('Failed to download %s to %s', remote_path, DATA_DIR) raise error.TestError('Failed to download file %s from bucket.' % file) return os.path.join(DATA_DIR, file) def generate_noisy_speech_file(self, speech_path, noise_path): """Generate the mixed audio file of speech and noise data. @param speech_path: the file path of the pure speech audio. @param noise_path: the file path of the noise audio. @raises: error.TestError if failed. @returns: the file path of the mixed audio. """ mixed_wav_path = os.path.join(DATA_DIR, 'speech_noise_mixed.wav') if os.path.exists(mixed_wav_path): os.remove(mixed_wav_path) sox_utils.mix_two_wav_files(speech_path, noise_path, mixed_wav_path, input_volume=1.0) if not os.path.isfile(mixed_wav_path): logging.error('WAV file %s does not exist.', mixed_wav_path) raise error.TestError('Failed to mix %s and %s by sox commands.' % (speech_path, noise_path)) return mixed_wav_path def run_once(self, test_data): """Runs Audio Noise Cancellation test. Test scenarios can be distinguished by the elements (keys) in test_data. Noisy environment test: test_data = dict( speech_file: the WAV file for the pure speech data. noise_file: the WAV file for the noise data. threshold: the min required score gain for NC effect.) Quiet environment test: test_data = dict( speech_file: the WAV file for the pure speech data. threshold: the min score diff tolerance for NC effect.) @param test_data: the dict for files and threshold as mentioned above. """ if not self.facade.get_noise_cancellation_supported(): logging.warning('Noise Cancellation is not supported.') raise error.TestWarn('Noise Cancellation is not supported.') def _remove_at_cleanup(filepath): self.cleanup_files.append(filepath) # Download the files from bucket. speech_path = self.download_file_from_bucket(test_data['speech_file']) _remove_at_cleanup(speech_path) ref_infos = sox_utils.get_infos_from_wav_file(speech_path) if ref_infos is None: raise error.TestError('Failed to get infos from wav file %s.' % speech_path) if 'noise_file' in test_data: # Noisy environment test when 'noise_file' is given. noise_path = self.download_file_from_bucket( test_data['noise_file']) _remove_at_cleanup(noise_path) test_audio_path = self.generate_noisy_speech_file( speech_path, noise_path) _remove_at_cleanup(test_audio_path) test_infos = sox_utils.get_infos_from_wav_file(test_audio_path) if test_infos is None: raise error.TestError('Failed to get infos from wav file %s.' % test_audio_path) else: # Quiet environment test. test_audio_path = speech_path test_infos = ref_infos playback_testdata = audio_test_data.AudioTestData( path=test_audio_path, data_format=dict(file_type='wav', sample_format='S{}_LE'.format( test_infos['bits']), channel=test_infos['channels'], rate=test_infos['rate']), duration_secs=test_infos['duration']) # Get and set VISQOL working environment. get_visqol_binary() # Bypass blocking mechanism in Cras to make sure Noise Cancellation is # enabled. self.facade.set_bypass_block_noise_cancellation(bypass=True) source = self.widget_factory.create_widget( chameleon_audio_ids.ChameleonIds.LINEOUT) sink = self.widget_factory.create_widget( chameleon_audio_ids.PeripheralIds.SPEAKER) binder = self.widget_factory.create_binder(source, sink) recorder = self.widget_factory.create_widget( chameleon_audio_ids.CrosIds.INTERNAL_MIC) # Select and check the node selected by cras is correct. audio_test_utils.check_and_set_chrome_active_node_types( self.facade, None, audio_test_utils.get_internal_mic_node(self.host)) # Adjust the proper input gain. self.facade.set_chrome_active_input_gain(50) # Stop Chrome UI to avoid NC state preference intervened by Chrome. self.host.run('stop ui') logging.info( 'UI is stopped to avoid NC preference intervention from Chrome' ) def _run_routine(recorded_filename, nc_enabled): # Set NC state via D-Bus control. self.facade.set_noise_cancellation_enabled(nc_enabled) time.sleep(self.DELAY_AFTER_NC_TOGGLED) with chameleon_audio_helper.bind_widgets(binder): time.sleep(self.DELAY_AFTER_BINDING) logfile_suffix = 'nc_on' if nc_enabled else 'nc_off' audio_test_utils.dump_cros_audio_logs( self.host, self.facade, self.resultsdir, 'after_binding.{}'.format(logfile_suffix)) logging.info('Set playback data on Chameleon') source.set_playback_data(playback_testdata) # Start recording, wait a few seconds, and then start playback. # Make sure the recorded data has silent samples in the # beginning to trim, and includes the entire playback content. logging.info('Start recording from Cros device') recorder.start_recording() time.sleep(self.DELAY_BEFORE_PLAYBACK_SECONDS) logging.info('Start playing %s from Chameleon', playback_testdata.path) source.start_playback() time.sleep(test_infos['duration'] + self.DELAY_AFTER_PLAYBACK_SECONDS) recorder.stop_recording() logging.info('Stopped recording from Cros device.') audio_test_utils.dump_cros_audio_logs( self.host, self.facade, self.resultsdir, 'after_recording.{}'.format(logfile_suffix)) recorder.read_recorded_binary() logging.info('Read recorded binary from Cros device.') # Remove the beginning of recorded data. This is to avoid artifact # caused by Cros device codec initialization in the beginning of # recording. recorder.remove_head(1.0) recorded_file = os.path.join(self.resultsdir, recorded_filename + '.raw') logging.info('Saving recorded data to %s', recorded_file) recorder.save_file(recorded_file) _remove_at_cleanup(recorded_file) # WAV file is also saved by recorder.save_file(). recorded_wav_path = recorded_file + '.wav' if not os.path.isfile(recorded_wav_path): logging.error('WAV file %s does not exist.', recorded_wav_path) raise error.TestError('Failed to find recorded wav file.') _remove_at_cleanup(recorded_wav_path) rec_infos = sox_utils.get_infos_from_wav_file(recorded_wav_path) if rec_infos is None: raise error.TestError('Failed to get infos from wav file %s.' % recorded_wav_path) # Downsample the recorded data from 48k to 16k rate. It is required # for getting ViSQOL score in speech mode. recorded_16k_path = '{}_16k{}'.format( *os.path.splitext(recorded_wav_path)) sox_utils.convert_format(recorded_wav_path, rec_infos['channels'], rec_infos['bits'], rec_infos['rate'], recorded_16k_path, ref_infos['channels'], ref_infos['bits'], ref_infos['rate'], 1.0, use_src_header=True, use_dst_header=True) # Remove the silence in the beginning and trim to the same duration # as the reference file. trimmed_recorded_16k_path = '{}_trim{}'.format( *os.path.splitext(recorded_16k_path)) sox_utils.trim_silence_from_wav_file(recorded_16k_path, trimmed_recorded_16k_path, ref_infos['duration'], duration_threshold=0.05) score = visqol_utils.get_visqol_score( ref_file=speech_path, deg_file=trimmed_recorded_16k_path, log_dir=self.resultsdir, speech_mode=True) logging.info('Recorded audio %s got ViSQOL score: %f', recorded_filename, score) return score logging.info('Run routine with NC enabled...') nc_on_score = _run_routine('record_nc_enabled', nc_enabled=True) logging.info('Run routine with NC disabled...') nc_off_score = _run_routine('record_nc_disabled', nc_enabled=False) score_diff = nc_on_score - nc_off_score # Track ViSQOL performance score test_desc = 'internal_mic_noise_cancellation_visqol_diff' self.write_perf_keyval({test_desc: score_diff}) if score_diff < test_data['threshold']: raise error.TestFail( 'ViSQOL score diff for NC(=%f) is lower than threshold(=%f)' % (score_diff, test_data['threshold']))