def test_close_temp_files():
'''
Create a bunch of temp files and then make sure they've been closed and
deleted.
'''
# With delete=True
tmpfiles = []
with _close_temp_files(tmpfiles):
for _ in range(5):
tmpfiles.append(
tempfile.NamedTemporaryFile(suffix='.wav', delete=True))
for tf in tmpfiles:
assert tf.file.closed
assert not os.path.isfile(tf.name)
# With delete=False
tmpfiles = []
with _close_temp_files(tmpfiles):
for _ in range(5):
tmpfiles.append(
tempfile.NamedTemporaryFile(suffix='.wav', delete=False))
for tf in tmpfiles:
assert tf.file.closed
assert not os.path.isfile(tf.name)
def test_match_sample_length():
durations_to_match = [1, 2, 5, 7, 22500, 44100, 88200, 100001]
invalid_durations_to_match = [0, -1, .5, 1.0]
tmpfiles = []
with _close_temp_files(tmpfiles):
carhorn = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
shutil.copyfile(CARHORN_FILE, carhorn.name)
tmpfiles.append(carhorn)
siren = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
shutil.copyfile(SIREN_FILE, siren.name)
tmpfiles.append(siren)
for _duration in durations_to_match:
match_sample_length(carhorn.name, _duration)
carhorn_audio, _ = sf.read(carhorn.name)
assert carhorn_audio.shape[0] == _duration
match_sample_length(siren.name, _duration)
siren_audio, _ = sf.read(siren.name)
assert siren_audio.shape[0] == _duration
# should be summable
summed_events = sum([carhorn_audio, siren_audio])
assert summed_events.shape[0] == _duration
for _duration in invalid_durations_to_match:
pytest.raises(ScaperError, match_sample_length, carhorn.name, _duration)
pytest.raises(ScaperError, match_sample_length, siren.name, _duration)
def test_close_temp_files():
'''
Create a bunch of temp files and then make sure they've been closed and
deleted.
'''
# With delete=True
tmpfiles = []
with _close_temp_files(tmpfiles):
for _ in range(5):
tmpfiles.append(
tempfile.NamedTemporaryFile(suffix='.wav', delete=True))
for tf in tmpfiles:
assert tf.file.closed
assert not os.path.isfile(tf.name)
# With delete=False
tmpfiles = []
with _close_temp_files(tmpfiles):
for _ in range(5):
tmpfiles.append(
tempfile.NamedTemporaryFile(suffix='.wav', delete=False))
for tf in tmpfiles:
assert tf.file.closed
assert not os.path.isfile(tf.name)
# with an exception before exiting
try:
tmpfiles = []
with _close_temp_files(tmpfiles):
tmpfiles.append(
tempfile.NamedTemporaryFile(suffix='.wav', delete=True))
raise ScaperError
except ScaperError:
for tf in tmpfiles:
assert tf.file.closed
assert not os.path.isfile(tf.name)
else:
assert False, 'Exception was not reraised.'
def test_close_temp_files():
'''
Create a bunch of temp files and then make sure they've been closed and
deleted.
'''
# With delete=True
tmpfiles = []
with _close_temp_files(tmpfiles):
for _ in range(5):
tmpfiles.append(
tempfile.NamedTemporaryFile(suffix='.wav', delete=True))
for tf in tmpfiles:
assert tf.file.closed
assert not os.path.isfile(tf.name)
# With delete=False
tmpfiles = []
with _close_temp_files(tmpfiles):
for _ in range(5):
tmpfiles.append(
tempfile.NamedTemporaryFile(suffix='.wav', delete=False))
for tf in tmpfiles:
assert tf.file.closed
assert not os.path.isfile(tf.name)
# with an exception before exiting
try:
tmpfiles = []
with _close_temp_files(tmpfiles):
tmpfiles.append(
tempfile.NamedTemporaryFile(suffix='.wav', delete=True))
raise ScaperError
except ScaperError:
for tf in tmpfiles:
assert tf.file.closed
assert not os.path.isfile(tf.name)
else:
assert False, 'Exception was not reraised.'
def test_generate_audio(atol=1e-4, rtol=1e-8):
# Regression test: same spec, same audio (not this will fail if we update
# any of the audio processing techniques used (e.g. change time stretching
# algorithm.
sc = scaper.Scaper(10.0, fg_path=FG_PATH, bg_path=BG_PATH)
sc.ref_db = -50
# background
sc.add_background(
label=('const', 'park'),
source_file=('const', 'tests/data/audio/background/park/'
'268903__yonts__city-park-tel-aviv-israel.wav'),
source_time=('const', 0))
# foreground events
sc.add_event(label=('const', 'siren'),
source_file=('const', 'tests/data/audio/foreground/'
'siren/69-Siren-1.wav'),
source_time=('const', 5),
event_time=('const', 2),
event_duration=('const', 5),
snr=('const', 5),
pitch_shift=None,
time_stretch=None)
sc.add_event(label=('const', 'car_horn'),
source_file=('const', 'tests/data/audio/foreground/'
'car_horn/17-CAR-Rolls-Royce-Horn.wav'),
source_time=('const', 0),
event_time=('const', 5),
event_duration=('const', 2),
snr=('const', 20),
pitch_shift=('const', 1),
time_stretch=None)
sc.add_event(
label=('const', 'human_voice'),
source_file=('const', 'tests/data/audio/foreground/'
'human_voice/42-Human-Vocal-Voice-taxi-2_edit.wav'),
source_time=('const', 0),
event_time=('const', 7),
event_duration=('const', 2),
snr=('const', 10),
pitch_shift=None,
time_stretch=('const', 1.2))
tmpfiles = []
with _close_temp_files(tmpfiles):
wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
tmpfiles.append(wav_file)
jam = sc._instantiate(disable_instantiation_warnings=True)
sc._generate_audio(wav_file.name, jam.annotations[0])
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
# with reverb
sc._generate_audio(wav_file.name, jam.annotations[0], reverb=0.2)
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_REVERB_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
# Don't disable sox warnings (just to cover line)
sc._generate_audio(wav_file.name,
jam.annotations[0],
disable_sox_warnings=False)
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
# namespace must be scaper
jam.annotations[0].namespace = 'tag_open'
pytest.raises(ScaperError, sc._generate_audio, wav_file.name,
jam.annotations[0])
# unsupported event role must raise error
jam.annotations[0].namespace = 'scaper'
jam.annotations[0].data[3].value['role'] = 'ewok'
pytest.raises(ScaperError, sc._generate_audio, wav_file.name,
jam.annotations[0])
# soundscape with no events will raise warning and won't generate audio
sc = scaper.Scaper(10.0, fg_path=FG_PATH, bg_path=BG_PATH)
sc.ref_db = -50
jam = sc._instantiate(disable_instantiation_warnings=True)
pytest.warns(ScaperWarning, sc._generate_audio, wav_file.name,
jam.annotations[0])
# soundscape with only one event will use transformer (regression test)
sc = scaper.Scaper(10.0, fg_path=FG_PATH, bg_path=BG_PATH)
sc.ref_db = -20
# background
sc.add_background(
label=('const', 'park'),
source_file=('const', 'tests/data/audio/background/park/'
'268903__yonts__city-park-tel-aviv-israel.wav'),
source_time=('const', 0))
jam = sc._instantiate(disable_instantiation_warnings=True)
sc._generate_audio(wav_file.name, jam.annotations[0], reverb=0.2)
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_BGONLY_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
def test_trim(atol=1e-5, rtol=1e-8):
# Things we want to test:
# 1. Jam trimmed correctly (mainly handled by jams.slice)
# 2. value dict updated correctly (event_time, event_duration, source_time)
# 3. scaper sandbox updated correctly (n_events, poly, gini, duration)
# 4. audio trimmed correctly
tmpfiles = []
with _close_temp_files(tmpfiles):
# Create all necessary temp files
orig_wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
orig_jam_file = tempfile.NamedTemporaryFile(suffix='.jams',
delete=True)
trim_wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
trim_jam_file = tempfile.NamedTemporaryFile(suffix='.jams',
delete=True)
trimstrict_wav_file = tempfile.NamedTemporaryFile(suffix='.wav',
delete=True)
trimstrict_jam_file = tempfile.NamedTemporaryFile(suffix='.jams',
delete=True)
tmpfiles.append(orig_wav_file)
tmpfiles.append(orig_jam_file)
tmpfiles.append(trim_wav_file)
tmpfiles.append(trim_jam_file)
tmpfiles.append(trimstrict_wav_file)
tmpfiles.append(trimstrict_jam_file)
# --- Create soundscape and save to tempfiles --- #
sc = scaper.Scaper(10, FG_PATH, BG_PATH)
sc.protected_labels = []
sc.ref_db = -50
sc.add_background(label=('const', 'park'),
source_file=('choose', []),
source_time=('const', 0))
# Add 5 events
start_times = [0.5, 2.5, 4.5, 6.5, 8.5]
for event_time in start_times:
sc.add_event(label=('const', 'siren'),
source_file=('choose', []),
source_time=('const', 5),
event_time=('const', event_time),
event_duration=('const', 1),
snr=('const', 10),
pitch_shift=None,
time_stretch=None)
sc.generate(orig_wav_file.name,
orig_jam_file.name,
disable_instantiation_warnings=True)
# --- Trim soundscape using scaper.trim with strict=False --- #
scaper.trim(orig_wav_file.name,
orig_jam_file.name,
trim_wav_file.name,
trim_jam_file.name,
3,
7,
no_audio=False)
# --- Validate output --- #
# validate JAMS
trimjam = jams.load(trim_jam_file.name)
trimann = trimjam.annotations.search(namespace='scaper')[0]
# Time and duration of annotation observation must be changed, but
# values in the value dict must remained unchanged!
for event in trimann.data:
if event.value['role'] == 'background':
assert (event.time == 0 and event.duration == 4
and event.value['event_time'] == 0
and event.value['event_duration'] == 10
and event.value['source_time'] == 0)
else:
if event.time == 0:
assert (event.duration == 0.5
and event.value['event_time'] == 2.5
and event.value['event_duration'] == 1
and event.value['source_time'] == 5)
elif event.time == 1.5:
assert (event.duration == 1
and event.value['event_time'] == 4.5
and event.value['event_duration'] == 1
and event.value['source_time'] == 5)
elif event.time == 3.5:
assert (event.duration == 0.5
and event.value['event_time'] == 6.5
and event.value['event_duration'] == 1
and event.value['source_time'] == 5)
else:
assert False
# validate audio
orig_wav, sr = soundfile.read(orig_wav_file.name)
trim_wav, sr = soundfile.read(trim_wav_file.name)
assert np.allclose(trim_wav,
orig_wav[3 * sr:7 * sr],
atol=atol,
rtol=rtol)
def test_generate_from_jams(atol=1e-5, rtol=1e-8):
# Test for invalid jams: no annotations
tmpfiles = []
with _close_temp_files(tmpfiles):
jam = jams.JAMS()
jam.file_metadata.duration = 10
jam_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
gen_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
jam.save(jam_file.name)
pytest.raises(ScaperError, scaper.generate_from_jams, jam_file.name,
gen_file.name)
# Test for valid jams files
tmpfiles = []
with _close_temp_files(tmpfiles):
# Create all necessary temp files
orig_wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
orig_jam_file = tempfile.NamedTemporaryFile(suffix='.jams',
delete=True)
gen_wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
gen_jam_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
tmpfiles.append(orig_wav_file)
tmpfiles.append(orig_jam_file)
tmpfiles.append(gen_wav_file)
tmpfiles.append(gen_jam_file)
# --- Define scaper --- *
sc = scaper.Scaper(10, FG_PATH, BG_PATH)
sc.protected_labels = []
sc.ref_db = -50
sc.add_background(label=('choose', []),
source_file=('choose', []),
source_time=('const', 0))
# Add 5 events
for _ in range(5):
sc.add_event(label=('choose', []),
source_file=('choose', []),
source_time=('const', 0),
event_time=('uniform', 0, 9),
event_duration=('choose', [1, 2, 3]),
snr=('uniform', 10, 20),
pitch_shift=('uniform', -1, 1),
time_stretch=('uniform', 0.8, 1.2))
# generate, then generate from the jams and compare audio files
# repeat 5 time
for _ in range(5):
sc.generate(orig_wav_file.name,
orig_jam_file.name,
disable_instantiation_warnings=True)
scaper.generate_from_jams(orig_jam_file.name, gen_wav_file.name)
# validate audio
orig_wav, sr = soundfile.read(orig_wav_file.name)
gen_wav, sr = soundfile.read(gen_wav_file.name)
assert np.allclose(gen_wav, orig_wav, atol=atol, rtol=rtol)
# Now add in trimming!
for _ in range(5):
sc.generate(orig_wav_file.name,
orig_jam_file.name,
disable_instantiation_warnings=True)
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 5), np.random.uniform(5, 10))
scaper.generate_from_jams(orig_jam_file.name, gen_wav_file.name)
# validate audio
orig_wav, sr = soundfile.read(orig_wav_file.name)
gen_wav, sr = soundfile.read(gen_wav_file.name)
assert np.allclose(gen_wav, orig_wav, atol=atol, rtol=rtol)
# Double trimming
for _ in range(2):
sc.generate(orig_wav_file.name,
orig_jam_file.name,
disable_instantiation_warnings=True)
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 2), np.random.uniform(8, 10))
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 2), np.random.uniform(4, 6))
scaper.generate_from_jams(orig_jam_file.name, gen_wav_file.name)
# Tripple trimming
for _ in range(2):
sc.generate(orig_wav_file.name,
orig_jam_file.name,
disable_instantiation_warnings=True)
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 2), np.random.uniform(8, 10))
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 1), np.random.uniform(5, 6))
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 1), np.random.uniform(3, 4))
scaper.generate_from_jams(orig_jam_file.name, gen_wav_file.name)
# validate audio
orig_wav, sr = soundfile.read(orig_wav_file.name)
gen_wav, sr = soundfile.read(gen_wav_file.name)
assert np.allclose(gen_wav, orig_wav, atol=atol, rtol=rtol)
# Test with new FG and BG paths
for _ in range(5):
sc.generate(orig_wav_file.name,
orig_jam_file.name,
disable_instantiation_warnings=True)
scaper.generate_from_jams(orig_jam_file.name,
gen_wav_file.name,
fg_path=ALT_FG_PATH,
bg_path=ALT_BG_PATH)
# validate audio
orig_wav, sr = soundfile.read(orig_wav_file.name)
gen_wav, sr = soundfile.read(gen_wav_file.name)
assert np.allclose(gen_wav, orig_wav, atol=atol, rtol=rtol)
# Ensure jam file saved correctly
scaper.generate_from_jams(orig_jam_file.name,
gen_wav_file.name,
jams_outfile=gen_jam_file.name)
orig_jam = jams.load(orig_jam_file.name)
gen_jam = jams.load(gen_jam_file.name)
assert orig_jam == gen_jam
def test_generate(atol=1e-4, rtol=1e-8):
# Final regression test on all files
sc = scaper.Scaper(10.0, fg_path=FG_PATH, bg_path=BG_PATH)
sc.ref_db = -50
# background
sc.add_background(
label=('const', 'park'),
source_file=('const', 'tests/data/audio/background/park/'
'268903__yonts__city-park-tel-aviv-israel.wav'),
source_time=('const', 0))
# foreground events
sc.add_event(label=('const', 'siren'),
source_file=('const', 'tests/data/audio/foreground/'
'siren/69-Siren-1.wav'),
source_time=('const', 5),
event_time=('const', 2),
event_duration=('const', 5),
snr=('const', 5),
pitch_shift=None,
time_stretch=None)
sc.add_event(label=('const', 'car_horn'),
source_file=('const', 'tests/data/audio/foreground/'
'car_horn/17-CAR-Rolls-Royce-Horn.wav'),
source_time=('const', 0),
event_time=('const', 5),
event_duration=('const', 2),
snr=('const', 20),
pitch_shift=('const', 1),
time_stretch=None)
sc.add_event(
label=('const', 'human_voice'),
source_file=('const', 'tests/data/audio/foreground/'
'human_voice/42-Human-Vocal-Voice-taxi-2_edit.wav'),
source_time=('const', 0),
event_time=('const', 7),
event_duration=('const', 2),
snr=('const', 10),
pitch_shift=None,
time_stretch=('const', 1.2))
tmpfiles = []
with _close_temp_files(tmpfiles):
wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
jam_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
txt_file = tempfile.NamedTemporaryFile(suffix='.txt', delete=True)
tmpfiles.append(wav_file)
tmpfiles.append(jam_file)
tmpfiles.append(txt_file)
sc.generate(wav_file.name,
jam_file.name,
txt_path=txt_file.name,
disable_instantiation_warnings=True)
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
# validate jams
jam = jams.load(jam_file.name)
regjam = jams.load(REG_JAM_PATH)
_compare_scaper_jams(jam, regjam)
# validate txt
# read in both files
txt_data = []
with open(txt_file.name) as file:
reader = csv.reader(file, delimiter='\t')
for row in reader:
txt_data.append(row)
txt_data = np.asarray(txt_data)
regtxt_data = []
with open(REG_TXT_PATH) as file:
reader = csv.reader(file, delimiter='\t')
for row in reader:
regtxt_data.append(row)
regtxt_data = np.asarray(regtxt_data)
# compare start and end times
assert np.allclose([float(x) for x in txt_data[:, 0]],
[float(x) for x in regtxt_data[:, 0]])
assert np.allclose([float(x) for x in txt_data[:, 1]],
[float(x) for x in regtxt_data[:, 1]])
# compare labels
assert (txt_data[:, 2] == regtxt_data[:, 2]).all()
# reverb value must be in (0, 1) range
for reverb in [-1, 2]:
pytest.raises(ScaperError,
sc.generate,
wav_file.name,
jam_file.name,
reverb=reverb,
disable_instantiation_warnings=True)
def _test_generate_audio(SR, REG_WAV_PATH, REG_BGONLY_WAV_PATH, REG_REVERB_WAV_PATH, atol=1e-4, rtol=1e-8):
# Regression test: same spec, same audio (not this will fail if we update
# any of the audio processing techniques used (e.g. change time stretching
# algorithm.
sc = scaper.Scaper(10.0, fg_path=FG_PATH, bg_path=BG_PATH)
sc.ref_db = -50
sc.sr = SR
# background
sc.add_background(
label=('const', 'park'),
source_file=(
'const',
'tests/data/audio/background/park/'
'268903__yonts__city-park-tel-aviv-israel.wav'),
source_time=('const', 0))
# foreground events
sc.add_event(
label=('const', 'siren'),
source_file=('const',
'tests/data/audio/foreground/'
'siren/69-Siren-1.wav'),
source_time=('const', 5),
event_time=('const', 2),
event_duration=('const', 5),
snr=('const', 5),
pitch_shift=None,
time_stretch=None)
sc.add_event(
label=('const', 'car_horn'),
source_file=('const',
'tests/data/audio/foreground/'
'car_horn/17-CAR-Rolls-Royce-Horn.wav'),
source_time=('const', 0),
event_time=('const', 5),
event_duration=('const', 2),
snr=('const', 20),
pitch_shift=('const', 1),
time_stretch=None)
sc.add_event(
label=('const', 'human_voice'),
source_file=('const',
'tests/data/audio/foreground/'
'human_voice/42-Human-Vocal-Voice-taxi-2_edit.wav'),
source_time=('const', 0),
event_time=('const', 7),
event_duration=('const', 2),
snr=('const', 10),
pitch_shift=None,
time_stretch=('const', 1.2))
tmpfiles = []
with _close_temp_files(tmpfiles):
wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
tmpfiles.append(wav_file)
jam = sc._instantiate(disable_instantiation_warnings=True)
sc._generate_audio(wav_file.name, jam.annotations[0])
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
# with reverb
sc._generate_audio(wav_file.name, jam.annotations[0], reverb=0.2)
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_REVERB_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
# Don't disable sox warnings (just to cover line)
sc._generate_audio(wav_file.name, jam.annotations[0],
disable_sox_warnings=False)
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
# namespace must be scaper
jam.annotations[0].namespace = 'tag_open'
pytest.raises(ScaperError, sc._generate_audio, wav_file.name,
jam.annotations[0])
# unsupported event role must raise error
jam.annotations[0].namespace = 'scaper'
jam.annotations[0].data[3].value['role'] = 'ewok'
pytest.raises(ScaperError, sc._generate_audio, wav_file.name,
jam.annotations[0])
# soundscape with no events will raise warning and won't generate audio
sc = scaper.Scaper(10.0, fg_path=FG_PATH, bg_path=BG_PATH)
sc.ref_db = -50
jam = sc._instantiate(disable_instantiation_warnings=True)
pytest.warns(ScaperWarning, sc._generate_audio, wav_file.name,
jam.annotations[0])
# soundscape with only one event will use transformer (regression test)
sc = scaper.Scaper(10.0, fg_path=FG_PATH, bg_path=BG_PATH)
sc.ref_db = -20
sc.sr = SR
# background
sc.add_background(
label=('const', 'park'),
source_file=('const',
'tests/data/audio/background/park/'
'268903__yonts__city-park-tel-aviv-israel.wav'),
source_time=('const', 0))
reverb = 0.2
jam = sc._instantiate(disable_instantiation_warnings=True, reverb=reverb)
sc._generate_audio(wav_file.name, jam.annotations[0], reverb=reverb)
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_BGONLY_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
def test_trim(atol=1e-5, rtol=1e-8):
# Things we want to test:
# 1. Jam trimmed correctly (mainly handled by jams.slice)
# 2. value dict updated correctly (event_time, event_duration, source_time)
# 3. scaper sandbox updated correctly (n_events, poly, gini, duration)
# 4. audio trimmed correctly
tmpfiles = []
with _close_temp_files(tmpfiles):
# Create all necessary temp files
orig_wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
orig_jam_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
trim_wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
trim_jam_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
trimstrict_wav_file = tempfile.NamedTemporaryFile(
suffix='.wav', delete=True)
trimstrict_jam_file = tempfile.NamedTemporaryFile(
suffix='.jams', delete=True)
tmpfiles.append(orig_wav_file)
tmpfiles.append(orig_jam_file)
tmpfiles.append(trim_wav_file)
tmpfiles.append(trim_jam_file)
tmpfiles.append(trimstrict_wav_file)
tmpfiles.append(trimstrict_jam_file)
# --- Create soundscape and save to tempfiles --- #
sc = scaper.Scaper(10, FG_PATH, BG_PATH)
sc.protected_labels = []
sc.ref_db = -50
sc.add_background(label=('const', 'park'),
source_file=('choose', []),
source_time=('const', 0))
# Add 5 events
start_times = [0.5, 2.5, 4.5, 6.5, 8.5]
for event_time in start_times:
sc.add_event(label=('const', 'siren'),
source_file=('choose', []),
source_time=('const', 5),
event_time=('const', event_time),
event_duration=('const', 1),
snr=('const', 10),
pitch_shift=None,
time_stretch=None)
sc.generate(orig_wav_file.name, orig_jam_file.name,
disable_instantiation_warnings=True)
# --- Trim soundscape using scaper.trim with strict=False --- #
scaper.trim(orig_wav_file.name, orig_jam_file.name,
trim_wav_file.name, trim_jam_file.name,
3, 7, no_audio=False)
# --- Validate output --- #
# validate JAMS
trimjam = jams.load(trim_jam_file.name)
trimann = trimjam.annotations.search(namespace='scaper')[0]
# Time and duration of annotation observation must be changed, but
# values in the value dict must remained unchanged!
for event in trimann.data:
if event.value['role'] == 'background':
assert (event.time == 0 and
event.duration == 4 and
event.value['event_time'] == 0 and
event.value['event_duration'] == 10 and
event.value['source_time'] == 0)
else:
if event.time == 0:
assert (event.duration == 0.5 and
event.value['event_time'] == 2.5 and
event.value['event_duration'] == 1 and
event.value['source_time'] == 5)
elif event.time == 1.5:
assert (event.duration == 1 and
event.value['event_time'] == 4.5 and
event.value['event_duration'] == 1 and
event.value['source_time'] == 5)
elif event.time == 3.5:
assert (event.duration == 0.5 and
event.value['event_time'] == 6.5 and
event.value['event_duration'] == 1 and
event.value['source_time'] == 5)
else:
assert False
# validate audio
orig_wav, sr = soundfile.read(orig_wav_file.name)
trim_wav, sr = soundfile.read(trim_wav_file.name)
assert np.allclose(trim_wav, orig_wav[3*sr:7*sr], atol=atol, rtol=rtol)
def test_generate_from_jams(atol=1e-5, rtol=1e-8):
# Test for invalid jams: no annotations
tmpfiles = []
with _close_temp_files(tmpfiles):
jam = jams.JAMS()
jam.file_metadata.duration = 10
jam_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
gen_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
jam.save(jam_file.name)
pytest.raises(ScaperError, scaper.generate_from_jams, jam_file.name,
gen_file.name)
# Test for valid jams files
tmpfiles = []
with _close_temp_files(tmpfiles):
# Create all necessary temp files
orig_wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
orig_jam_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
gen_wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
gen_jam_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
tmpfiles.append(orig_wav_file)
tmpfiles.append(orig_jam_file)
tmpfiles.append(gen_wav_file)
tmpfiles.append(gen_jam_file)
# --- Define scaper --- *
sc = scaper.Scaper(10, FG_PATH, BG_PATH)
sc.protected_labels = []
sc.ref_db = -50
sc.add_background(label=('choose', []),
source_file=('choose', []),
source_time=('const', 0))
# Add 5 events
for _ in range(5):
sc.add_event(label=('choose', []),
source_file=('choose', []),
source_time=('const', 0),
event_time=('uniform', 0, 9),
event_duration=('choose', [1, 2, 3]),
snr=('uniform', 10, 20),
pitch_shift=('uniform', -1, 1),
time_stretch=('uniform', 0.8, 1.2))
# generate, then generate from the jams and compare audio files
# repeat 5 time
for _ in range(5):
sc.generate(orig_wav_file.name, orig_jam_file.name,
disable_instantiation_warnings=True)
scaper.generate_from_jams(orig_jam_file.name, gen_wav_file.name)
# validate audio
orig_wav, sr = soundfile.read(orig_wav_file.name)
gen_wav, sr = soundfile.read(gen_wav_file.name)
assert np.allclose(gen_wav, orig_wav, atol=atol, rtol=rtol)
# Now add in trimming!
for _ in range(5):
sc.generate(orig_wav_file.name, orig_jam_file.name,
disable_instantiation_warnings=True)
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 5), np.random.uniform(5, 10))
scaper.generate_from_jams(orig_jam_file.name, gen_wav_file.name)
# validate audio
orig_wav, sr = soundfile.read(orig_wav_file.name)
gen_wav, sr = soundfile.read(gen_wav_file.name)
assert np.allclose(gen_wav, orig_wav, atol=atol, rtol=rtol)
# Double trimming
for _ in range(2):
sc.generate(orig_wav_file.name, orig_jam_file.name,
disable_instantiation_warnings=True)
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 2), np.random.uniform(8, 10))
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 2), np.random.uniform(4, 6))
scaper.generate_from_jams(orig_jam_file.name, gen_wav_file.name)
# Tripple trimming
for _ in range(2):
sc.generate(orig_wav_file.name, orig_jam_file.name,
disable_instantiation_warnings=True)
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 2), np.random.uniform(8, 10))
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 1), np.random.uniform(5, 6))
scaper.trim(orig_wav_file.name, orig_jam_file.name,
orig_wav_file.name, orig_jam_file.name,
np.random.uniform(0, 1), np.random.uniform(3, 4))
scaper.generate_from_jams(orig_jam_file.name, gen_wav_file.name)
# validate audio
orig_wav, sr = soundfile.read(orig_wav_file.name)
gen_wav, sr = soundfile.read(gen_wav_file.name)
assert np.allclose(gen_wav, orig_wav, atol=atol, rtol=rtol)
# Test with new FG and BG paths
for _ in range(5):
sc.generate(orig_wav_file.name, orig_jam_file.name,
disable_instantiation_warnings=True)
scaper.generate_from_jams(orig_jam_file.name, gen_wav_file.name,
fg_path=ALT_FG_PATH, bg_path=ALT_BG_PATH)
# validate audio
orig_wav, sr = soundfile.read(orig_wav_file.name)
gen_wav, sr = soundfile.read(gen_wav_file.name)
assert np.allclose(gen_wav, orig_wav, atol=atol, rtol=rtol)
# Ensure jam file saved correctly
scaper.generate_from_jams(orig_jam_file.name, gen_wav_file.name,
jams_outfile=gen_jam_file.name)
orig_jam = jams.load(orig_jam_file.name)
gen_jam = jams.load(gen_jam_file.name)
assert orig_jam == gen_jam
def _test_generate(SR, REG_WAV_PATH, REG_JAM_PATH, REG_TXT_PATH, atol=1e-4, rtol=1e-8):
# Final regression test on all files
sc = scaper.Scaper(10.0, fg_path=FG_PATH, bg_path=BG_PATH)
sc.ref_db = -50
sc.sr = SR
# background
sc.add_background(
label=('const', 'park'),
source_file=(
'const',
'tests/data/audio/background/park/'
'268903__yonts__city-park-tel-aviv-israel.wav'),
source_time=('const', 0))
# foreground events
sc.add_event(
label=('const', 'siren'),
source_file=('const',
'tests/data/audio/foreground/'
'siren/69-Siren-1.wav'),
source_time=('const', 5),
event_time=('const', 2),
event_duration=('const', 5),
snr=('const', 5),
pitch_shift=None,
time_stretch=None)
sc.add_event(
label=('const', 'car_horn'),
source_file=('const',
'tests/data/audio/foreground/'
'car_horn/17-CAR-Rolls-Royce-Horn.wav'),
source_time=('const', 0),
event_time=('const', 5),
event_duration=('const', 2),
snr=('const', 20),
pitch_shift=('const', 1),
time_stretch=None)
sc.add_event(
label=('const', 'human_voice'),
source_file=('const',
'tests/data/audio/foreground/'
'human_voice/42-Human-Vocal-Voice-taxi-2_edit.wav'),
source_time=('const', 0),
event_time=('const', 7),
event_duration=('const', 2),
snr=('const', 10),
pitch_shift=None,
time_stretch=('const', 1.2))
tmpfiles = []
with _close_temp_files(tmpfiles):
wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
jam_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
txt_file = tempfile.NamedTemporaryFile(suffix='.txt', delete=True)
tmpfiles.append(wav_file)
tmpfiles.append(jam_file)
tmpfiles.append(txt_file)
sc.generate(wav_file.name, jam_file.name, txt_path=txt_file.name,
disable_instantiation_warnings=True)
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
# validate jams
jam = jams.load(jam_file.name)
regjam = jams.load(REG_JAM_PATH)
_compare_scaper_jams(jam, regjam)
# validate txt
# read in both files
txt_data = []
with open(txt_file.name) as file:
reader = csv.reader(file, delimiter='\t')
for row in reader:
txt_data.append(row)
txt_data = np.asarray(txt_data)
regtxt_data = []
with open(REG_TXT_PATH) as file:
reader = csv.reader(file, delimiter='\t')
for row in reader:
regtxt_data.append(row)
regtxt_data = np.asarray(regtxt_data)
# compare start and end times
assert np.allclose([float(x) for x in txt_data[:, 0]],
[float(x) for x in regtxt_data[:, 0]])
assert np.allclose([float(x) for x in txt_data[:, 1]],
[float(x) for x in regtxt_data[:, 1]])
# compare labels
assert (txt_data[:, 2] == regtxt_data[:, 2]).all()
# reverb value must be in (0, 1) range
for reverb in [-1, 2]:
pytest.raises(ScaperError, sc.generate, wav_file.name,
jam_file.name, reverb=reverb,
disable_instantiation_warnings=True)
def test_generate(atol=1e-4, rtol=1e-8):
# Final regression test on all files
sc = scaper.Scaper(10.0, fg_path=FG_PATH, bg_path=BG_PATH)
sc.ref_db = -50
# background
sc.add_background(
label=('const', 'park'),
source_file=('const', 'tests/data/audio/background/park/'
'268903__yonts__city-park-tel-aviv-israel.wav'),
source_time=('const', 0))
# foreground events
sc.add_event(label=('const', 'siren'),
source_file=('const', 'tests/data/audio/foreground/'
'siren/69-Siren-1.wav'),
source_time=('const', 5),
event_time=('const', 2),
event_duration=('const', 5),
snr=('const', 5),
pitch_shift=None,
time_stretch=None)
sc.add_event(label=('const', 'car_horn'),
source_file=('const', 'tests/data/audio/foreground/'
'car_horn/17-CAR-Rolls-Royce-Horn.wav'),
source_time=('const', 0),
event_time=('const', 5),
event_duration=('const', 2),
snr=('const', 20),
pitch_shift=('const', 1),
time_stretch=None)
sc.add_event(
label=('const', 'human_voice'),
source_file=('const', 'tests/data/audio/foreground/'
'human_voice/42-Human-Vocal-Voice-taxi-2_edit.wav'),
source_time=('const', 0),
event_time=('const', 7),
event_duration=('const', 2),
snr=('const', 10),
pitch_shift=None,
time_stretch=('const', 1.2))
tmpfiles = []
with _close_temp_files(tmpfiles):
wav_file = tempfile.NamedTemporaryFile(suffix='.wav', delete=True)
jam_file = tempfile.NamedTemporaryFile(suffix='.jams', delete=True)
txt_file = tempfile.NamedTemporaryFile(suffix='.txt', delete=True)
tmpfiles.append(wav_file)
tmpfiles.append(jam_file)
tmpfiles.append(txt_file)
sc.generate(wav_file.name,
jam_file.name,
txt_path=txt_file.name,
disable_instantiation_warnings=True)
# validate audio
wav, sr = soundfile.read(wav_file.name)
regwav, sr = soundfile.read(REG_WAV_PATH)
assert np.allclose(wav, regwav, atol=atol, rtol=rtol)
# validate jams
jam = jams.load(jam_file.name)
regjam = jams.load(REG_JAM_PATH)
# version might change, rest should be the same
regjam.annotations[0].sandbox.scaper['scaper_version'] = \
scaper.__version__
assert jam == regjam
# validate txt
txt = pd.read_csv(txt_file.name, header=None, sep='\t')
regtxt = pd.read_csv(REG_TXT_PATH, header=None, sep='\t')
assert (txt == regtxt).all().all()
# reverb value must be in (0, 1) range
for reverb in [-1, 2]:
pytest.raises(ScaperError,
sc.generate,
wav_file.name,
jam_file.name,
reverb=reverb,
disable_instantiation_warnings=True)
