def combine_rirs_and_sources(
rir_dataset,
source_dataset,
num_speakers,
dataset_name,
):
# The keys of rir_dataset are integers. Sort the rirs based on this
# integer.
rir_dataset = rir_dataset.sort(sort_fn=functools.partial(sorted, key=int))
assert len(rir_dataset) % len(source_dataset) == 0, (len(rir_dataset),
len(source_dataset))
repetitions = len(rir_dataset) // len(source_dataset)
source_dataset = source_dataset.sort()
source_dataset = list(source_dataset.tile(repetitions))
speaker_ids = [example['speaker_id'] for example in source_dataset]
rng = get_rng(dataset_name, 'example_compositions')
composition_examples = None
for _ in range(num_speakers):
composition_examples = extend_composition_example_greedy(
rng,
speaker_ids,
example_compositions=composition_examples,
)
ex_dict = dict()
assert len(rir_dataset) == len(composition_examples), (
len(rir_dataset), len(composition_examples))
for rir_example, composition_example in zip(rir_dataset,
composition_examples):
source_examples = [source_dataset[i] for i in composition_example]
example = get_randomized_example(
rir_example,
source_examples,
rng,
dataset_name,
)
ex_dict[example['example_id']] = example
return ex_dict
def main(json_path: Path, rir_dir: Path, wsj_json_path: Path, num_speakers):
wsj_json_path = Path(wsj_json_path).expanduser().resolve()
json_path = Path(json_path).expanduser().resolve()
if json_path.exists():
raise FileExistsError(json_path)
rir_dir = Path(rir_dir).expanduser().resolve()
assert wsj_json_path.is_file(), json_path
assert rir_dir.exists(), rir_dir
setup = dict(
train_si284=dict(source_dataset_name="train_si284"),
cv_dev93=dict(source_dataset_name="cv_dev93"),
test_eval92=dict(source_dataset_name="test_eval92"),
)
rir_db = JsonDatabase(rir_dir / "scenarios.json")
source_db = JsonDatabase(wsj_json_path)
target_db = dict()
target_db['datasets'] = defaultdict(dict)
for dataset_name in setup.keys():
source_dataset_name = setup[dataset_name]["source_dataset_name"]
source_iterator = source_db.get_dataset(source_dataset_name)
print(f'length of source {dataset_name}: {len(source_iterator)}')
source_iterator = source_iterator.filter(
filter_fn=filter_punctuation_pronunciation, lazy=False)
print(f'length of source {dataset_name}: {len(source_iterator)} '
'(after punctuation filter)')
rir_iterator = rir_db.get_dataset(dataset_name)
assert len(rir_iterator) % len(source_iterator) == 0, (
f'To avoid a bias towards certain utterance the len '
f'rir_iterator ({len(rir_iterator)}) should be an integer '
f'multiple of len source_iterator ({len(source_iterator)}).')
print(f'length of rir {dataset_name}: {len(rir_iterator)}')
probe_path = rir_dir / dataset_name / "0"
available_speaker_positions = len(list(probe_path.glob('h_*.wav')))
assert num_speakers <= available_speaker_positions, (
f'Requested {num_speakers} num_speakers, while found only '
f'{available_speaker_positions} rirs in {probe_path}.')
info = soundfile.info(str(rir_dir / dataset_name / "0" / "h_0.wav"))
sample_rate_rir = info.samplerate
ex_wsj = source_iterator.random_choice(1)[0]
info = soundfile.SoundFile(ex_wsj['audio_path']['observation'])
sample_rate_wsj = info.samplerate
assert sample_rate_rir == sample_rate_wsj, (sample_rate_rir,
sample_rate_wsj)
rir_iterator = rir_iterator.sort(
sort_fn=functools.partial(sorted, key=int))
source_iterator = source_iterator.sort()
assert len(rir_iterator) % len(source_iterator) == 0
repeats = len(rir_iterator) // len(source_iterator)
source_iterator = source_iterator.tile(repeats)
speaker_ids = [example['speaker_id'] for example in source_iterator]
rng = get_rng(dataset_name, 'example_compositions')
example_compositions = None
for _ in range(num_speakers):
example_compositions = extend_example_composition_greedy(
rng,
speaker_ids,
example_compositions=example_compositions,
)
ex_dict = dict()
assert len(rir_iterator) == len(example_compositions)
for rir_example, example_composition in zip(rir_iterator,
example_compositions):
source_examples = source_iterator[example_composition]
example_id = "_".join([
rir_example['example_id'],
*[ex["example_id"] for ex in source_examples],
])
rng = get_rng(dataset_name, example_id)
example = get_randomized_example(
rir_example,
source_examples,
rng,
dataset_name,
rir_dir,
)
ex_dict[example_id] = example
target_db['datasets'][dataset_name] = ex_dict
json_path.parent.mkdir(exist_ok=True, parents=True)
with json_path.open('w') as f:
json.dump(target_db, f, indent=2, ensure_ascii=False)
print(f'{json_path} written')
def get_randomized_example(rir_example, source_examples, rng, dataset_name):
example_id = "_".join([
rir_example['example_id'],
*[source_ex["example_id"] for source_ex in source_examples],
])
rng = get_rng(dataset_name, example_id)
example = copy(rir_example)
example['example_id'] = example_id
example['dataset'] = dataset_name
assert len(source_examples) <= len(example['source_position'][0])
num_speakers = len(source_examples)
# Remove unused source positions and rirs (i.e. the scenarios.json was
# maybe generated with more speakers)
example['source_position'] = [
v[:num_speakers] for v in example['source_position']
]
example['audio_path']['rir'] = example['audio_path']['rir'][:num_speakers]
example['num_speakers'] = num_speakers
example['speaker_id'] = [exa['speaker_id'] for exa in source_examples]
# asserts that no speaker_id is used twice
assert len(set(example['speaker_id'])) == example['num_speakers']
example["source_id"] = [exa['example_id'] for exa in source_examples]
for k in ('gender', 'kaldi_transcription'):
example[k] = [exa[k] for exa in source_examples]
example['log_weights'] = rng.uniform(0,
5,
size=(example['num_speakers'], ))
example['log_weights'] -= np.mean(example['log_weights'])
example['log_weights'] = example['log_weights'].tolist()
# This way, at least the first speaker can have proper alignments,
# all other speakers can not be used for ASR.
def _get_num_samples(num_samples):
if isinstance(num_samples, dict):
return num_samples['observation']
else:
return num_samples
example['num_samples'] = dict()
example['num_samples']['original_source'] = [
_get_num_samples(exa['num_samples']) for exa in source_examples
]
example['num_samples']['observation'] = max(
example['num_samples']['original_source'])
example["offset"] = []
for k in range(example['num_speakers']):
excess_samples = (example['num_samples']['observation'] -
example['num_samples']['original_source'][k])
assert excess_samples >= 0, excess_samples
example["offset"].append(rng.randint(0, excess_samples + 1))
example['audio_path']['original_source'] = [
exa['audio_path']['observation'] for exa in source_examples
]
# example['audio_path']['rir']: Already defined in rir_example.
return example