def inference(config: DictConfig):
device = check_envirionment(config.eval.use_cuda)
model = load_test_model(config.eval, device)
if config.eval.dataset == 'kspon':
vocab = KsponSpeechVocabulary(
f'../../../data/vocab/aihub_{config.eval.output_unit}_vocabs.csv', output_unit=config.eval.output_unit
)
elif config.eval.dataset == 'libri':
vocab = LibriSpeechVocabulary('../../../data/vocab/tokenizer.vocab', 'data/vocab/tokenizer.model')
else:
raise ValueError("Unsupported Dataset : {0}".format(config.eval.dataset))
audio_paths, transcripts = load_dataset(config.eval.transcripts_path)
testset = SpectrogramDataset(audio_paths=audio_paths, transcripts=transcripts,
sos_id=vocab.sos_id, eos_id=vocab.eos_id,
dataset_path=config.eval.dataset_path, config=config, spec_augment=False)
evaluator = Evaluator(
dataset=testset,
vocab=vocab,
batch_size=config.eval.batch_size,
device=device,
num_workers=config.eval.num_workers,
print_every=config.eval.print_every,
decode=config.eval.decode,
beam_size=config.eval.k,
)
evaluator.evaluate(model)
def inference(opt):
device = check_envirionment(opt.use_cuda)
model = load_test_model(opt, device)
if opt.dataset == 'kspon':
vocab = KsponSpeechVocabulary(
f'../data/vocab/aihub_{opt.output_unit}_vocabs.csv',
output_unit=opt.output_unit)
elif opt.dataset == 'libri':
vocab = LibriSpeechVocabulary('../data/vocab/tokenizer.vocab',
'../data/vocab/tokenizer.model')
else:
raise ValueError("Unsupported Dataset : {0}".format(opt.dataset))
audio_paths, transcripts = load_dataset(opt.transcripts_path)
testset = SpectrogramDataset(audio_paths=audio_paths,
transcripts=transcripts,
sos_id=vocab.sos_id,
eos_id=vocab.eos_id,
dataset_path=opt.dataset_path,
opt=opt,
spec_augment=False)
evaluator = Evaluator(testset, vocab, opt.batch_size, device,
opt.num_workers, opt.print_every, opt.decode, opt.k)
evaluator.evaluate(model)
def inference(opt):
device = check_envirionment(opt.use_cuda)
model = load_test_model(opt, device)
audio_paths, script_paths = load_data_list(opt.data_list_path, opt.dataset_path)
target_dict = load_targets(script_paths)
testset = SpectrogramDataset(audio_paths=audio_paths, script_paths=script_paths, sos_id=SOS_token, eos_id=EOS_token,
target_dict=target_dict, opt=opt, spec_augment=False, noise_augment=False)
evaluator = Evaluator(testset, opt.batch_size, device, opt.num_workers, opt.print_every, opt.decode, opt.k)
evaluator.evaluate(model)
mfcc = mfcc[:, ::-1]
mfcc = torch.FloatTensor(np.ascontiguousarray(np.swapaxes(mfcc, 0, 1)))
return mfcc
parser = argparse.ArgumentParser(description='Run Pretrain')
parser.add_argument('--model_path', type=str, default='../pretrain/model.pt')
parser.add_argument('--audio_path',
type=str,
default='../pretrain/sample_audio.pcm')
parser.add_argument('--device', type=str, default='cuda')
opt = parser.parse_args()
feature_vector = parse_audio(opt.audio_path, del_silence=True)
input_length = torch.IntTensor([len(feature_vector)])
vocab = KsponSpeechVocabulary('../data/vocab/aihub_character_vocabs.csv')
model = load_test_model(opt, opt.device)
model.eval()
output = model(inputs=feature_vector.unsqueeze(0),
input_lengths=input_length,
teacher_forcing_ratio=0.0,
return_decode_dict=False)
logit = torch.stack(output, dim=1).to(opt.device)
pred = logit.max(-1)[1]
sentence = vocab.label_to_string(pred.cpu().detach().numpy())
print(sentence)