def recording_set():
return RecordingSet.from_recordings([
Recording(id='x',
sources=[
AudioSource(type='file',
channels=[0],
source='text/fixtures/mono_c0.wav'),
AudioSource(type='command',
channels=[1],
source='cat text/fixtures/mono_c1.wav')
],
sampling_rate=8000,
num_samples=4000,
duration=0.5)
])
def recording_set():
return RecordingSet.from_recordings([
Recording(
id="x",
sources=[
AudioSource(type="file",
channels=[0],
source="text/fixtures/mono_c0.wav"),
AudioSource(
type="command",
channels=[1],
source="cat text/fixtures/mono_c1.wav",
),
],
sampling_rate=8000,
num_samples=4000,
duration=0.5,
)
])
def dummy_recording(unique_id: int) -> Recording:
return Recording(id=f'dummy-recording-{unique_id:04d}',
sources=[
AudioSource(type='command',
channels=[0],
source='echo "dummy waveform"')
],
sampling_rate=16000,
num_samples=16000,
duration=1.0)
def with_recording(self, sampling_rate: int,
num_samples: int) -> Recording:
f = NamedTemporaryFile('wb', suffix='.wav')
self.files.append(f)
duration = num_samples / sampling_rate
samples = np.random.rand(num_samples)
soundfile.write(f.name, samples, samplerate=sampling_rate)
return Recording(
id=str(uuid4()),
sources=[AudioSource(type='file', channels=[0], source=f.name)],
sampling_rate=sampling_rate,
num_samples=num_samples,
duration=duration)
def make_recording(sampling_rate: int, num_samples: int) -> Recording:
# The idea is that we're going to write to a temporary file with a sine wave recording
# of specified duration and sampling rate, and clean up only after the test is executed.
with NamedTemporaryFile('wb', suffix='.wav') as f:
duration = num_samples / sampling_rate
samples: np.ndarray = np.sin(2 * np.pi * np.arange(0, num_samples) /
sampling_rate)
soundfile.write(f, samples, samplerate=sampling_rate)
yield Recording(
id=f'recording-{sampling_rate}-{duration}',
sources=[AudioSource(type='file', channels=[0], source=f.name)],
sampling_rate=sampling_rate,
num_samples=num_samples,
duration=duration)
def cut_set():
cut = MonoCut(
id="cut-1",
start=0.0,
duration=10.0,
channel=0,
features=Features(
type="fbank",
num_frames=100,
num_features=40,
frame_shift=0.01,
sampling_rate=16000,
start=0.0,
duration=10.0,
storage_type="lilcom",
storage_path="irrelevant",
storage_key="irrelevant",
),
recording=Recording(
id="rec-1",
sampling_rate=16000,
num_samples=160000,
duration=10.0,
sources=[
AudioSource(type="file", channels=[0], source="irrelevant")
],
),
supervisions=[
SupervisionSegment(id="sup-1",
recording_id="irrelevant",
start=0.5,
duration=6.0),
SupervisionSegment(id="sup-2",
recording_id="irrelevant",
start=7.0,
duration=2.0),
],
)
return CutSet.from_cuts([
cut,
fastcopy(cut, id="cut-nosup", supervisions=[]),
fastcopy(cut, id="cut-norec", recording=None),
fastcopy(cut, id="cut-nofeat", features=None),
cut.pad(duration=30.0, direction="left"),
cut.pad(duration=30.0, direction="right"),
cut.pad(duration=30.0, direction="both"),
cut.mix(cut, offset_other_by=5.0, snr=8),
])
def cut_set():
cut = Cut(id='cut-1',
start=0.0,
duration=10.0,
channel=0,
features=Features(
type='fbank',
num_frames=100,
num_features=40,
frame_shift=0.01,
sampling_rate=16000,
start=0.0,
duration=10.0,
storage_type='lilcom',
storage_path='irrelevant',
storage_key='irrelevant',
),
recording=Recording(id='rec-1',
sampling_rate=16000,
num_samples=160000,
duration=10.0,
sources=[
AudioSource(type='file',
channels=[0],
source='irrelevant')
]),
supervisions=[
SupervisionSegment(id='sup-1',
recording_id='irrelevant',
start=0.5,
duration=6.0),
SupervisionSegment(id='sup-2',
recording_id='irrelevant',
start=7.0,
duration=2.0)
])
return CutSet.from_cuts([
cut,
fastcopy(cut, id='cut-nosup', supervisions=[]),
fastcopy(cut, id='cut-norec', recording=None),
fastcopy(cut, id='cut-nofeat', features=None),
cut.pad(duration=30.0, direction='left'),
cut.pad(duration=30.0, direction='right'),
cut.pad(duration=30.0, direction='both'),
cut.mix(cut, offset_other_by=5.0, snr=8)
])
def with_recording(
self, sampling_rate: int, num_samples: int, use_zeros: bool = False
) -> Recording:
import torchaudio # torchaudio does not have issues on M1 macs unlike soundfile
f = NamedTemporaryFile("wb", suffix=".wav")
self.files.append(f)
duration = num_samples / sampling_rate
if use_zeros:
samples = torch.zeros((1, num_samples))
else:
samples = torch.rand((1, num_samples))
torchaudio.save(f.name, samples, sample_rate=sampling_rate)
f.flush()
os.fsync(f)
return Recording(
id=str(uuid4()),
sources=[AudioSource(type="file", channels=[0], source=f.name)],
sampling_rate=sampling_rate,
num_samples=num_samples,
duration=duration,
)
def make_recording_callhome(
sph_path: Pathlike,
recording_id: Optional[str] = None,
relative_path_depth: Optional[int] = None,
sph2pipe_path: Optional[Pathlike] = None
) -> Recording:
"""
This function creates manifests for CallHome recordings that are compressed
with shorten, a rare and mostly unsupported codec. You will need to install
sph2pipe (e.g. using Kaldi) in order to read these files.
"""
try:
from sphfile import SPHFile
except ImportError:
raise ImportError("Please install sphfile (pip install sphfile) instead and "
"try preparing CallHome English again.")
if sph2pipe_path is None:
sph2pipe_path = 'sph2pipe'
else:
sph2pipe_path = str(sph2pipe_path).strip()
sph_path = Path(sph_path)
sphf = SPHFile(sph_path)
return Recording(
id=recording_id if recording_id is not None else sph_path.stem,
sampling_rate=sphf.format['sample_rate'],
num_samples=sphf.format['sample_count'],
duration=sphf.format['sample_count'] / sphf.format['sample_rate'],
sources=[
AudioSource(
type='command',
channels=list(range(sphf.format['channel_count'])),
source=f'{sph2pipe_path} -f wav -p ' + (
'/'.join(sph_path.parts[-relative_path_depth:])
if relative_path_depth is not None and relative_path_depth > 0
else str(sph_path)
)
)
]
)
def file_source():
return AudioSource(type='file',
channels=[0],
source='test/fixtures/mono_c0.wav')
def file_source():
return AudioSource(type="file",
channels=[0],
source="test/fixtures/mono_c0.wav")
