def test_common_transducer(enc_type, enc_kwargs):
nnet_cls = aps_asr_nnet("asr@transducer")
vocab_size = 100
batch_size = 4
dec_kwargs = {
"embed_size": 512,
"enc_dim": 512,
"jot_dim": 512,
"dec_rnn": "lstm",
"dec_layers": 2,
"dec_hidden": 512,
"dec_dropout": 0.1
}
asr_transform = AsrTransform(feats="fbank-log-cmvn",
frame_len=400,
frame_hop=160,
window="hamm")
xfmr_encoders = ["xfmr_abs", "xfmr_rel", "xfmr_xl", "cfmr_xl"]
rnnt = nnet_cls(input_size=80,
vocab_size=vocab_size,
blank=vocab_size - 1,
asr_transform=asr_transform,
enc_type=enc_type,
enc_proj=None if enc_type in xfmr_encoders else 512,
enc_kwargs=enc_kwargs,
dec_kwargs=dec_kwargs)
x, x_len, y, y_len, u = gen_egs(vocab_size, batch_size)
z, _ = rnnt(x, x_len, y, y_len)
assert z.shape[2:] == th.Size([u + 1, vocab_size])
def test_xfmr_transducer(enc_type, enc_kwargs):
nnet_cls = aps_asr_nnet("asr@xfmr_transducer")
vocab_size = 100
batch_size = 4
dec_kwargs = {
"jot_dim": 512,
"att_dim": 512,
"nhead": 8,
"feedforward_dim": 2048,
"pos_dropout": 0.1,
"att_dropout": 0.1,
"num_layers": 2
}
asr_transform = AsrTransform(feats="fbank-log-cmvn",
frame_len=400,
frame_hop=160,
window="hamm")
xfmr_rnnt = nnet_cls(input_size=80,
vocab_size=vocab_size,
blank=vocab_size - 1,
asr_transform=asr_transform,
enc_type=enc_type,
enc_proj=512,
enc_kwargs=enc_kwargs,
dec_kwargs=dec_kwargs)
x, x_len, y, y_len, u = gen_egs(vocab_size, batch_size)
z, _ = xfmr_rnnt(x, x_len, y, y_len)
assert z.shape[2:] == th.Size([u + 1, vocab_size])
def test_att_encoder(enc_type, enc_kwargs):
nnet_cls = aps_asr_nnet("asr@att")
vocab_size = 100
batch_size = 4
asr_transform = AsrTransform(feats="fbank-log-cmvn",
frame_len=400,
frame_hop=160,
window="hamm")
att_asr = nnet_cls(input_size=80,
vocab_size=vocab_size,
sos=0,
eos=1,
ctc=True,
asr_transform=asr_transform,
att_type="ctx",
att_kwargs={"att_dim": 512},
enc_type=enc_type,
enc_proj=256,
enc_kwargs=enc_kwargs,
dec_type="rnn",
dec_dim=512,
dec_kwargs=default_rnn_dec_kwargs)
x, x_len, y, y_len, u = gen_egs(vocab_size, batch_size)
z, _, _, _ = att_asr(x, x_len, y, y_len)
assert z.shape == th.Size([4, u + 1, vocab_size - 1])
def test_asr_transform(wav, feats, shape):
transform = AsrTransform(feats=feats,
frame_len=400,
frame_hop=160,
use_power=True,
pre_emphasis=0.96,
aug_prob=0.5,
aug_mask_zero=False)
feats, _ = transform(th.from_numpy(wav[None, ...]), None)
assert feats.shape == th.Size(shape)
assert th.sum(th.isnan(feats)) == 0
assert transform.feats_dim == shape[-1]
def test_mvdr_att(att_type, att_kwargs):
nnet_cls = aps_asr_nnet("asr@enh_att")
vocab_size = 100
batch_size = 4
num_channels = 4
enh_kwargs = {
"rnn": "lstm",
"num_layers": 2,
"rnn_inp_proj": 512,
"hidden_size": 512,
"dropout": 0.2,
"bidirectional": False,
"mvdr_att_dim": 512,
"mask_norm": True,
"num_bins": 257
}
asr_transform = AsrTransform(feats="abs-mel-log-cmvn",
frame_len=400,
frame_hop=160,
window="hamm")
enh_transform = EnhTransform(feats="spectrogram-log-cmvn-ipd",
frame_len=400,
frame_hop=160,
window="hamm",
ipd_index="0,1;0,2;0,3",
cos_ipd=True)
mvdr_att_asr = nnet_cls(enh_input_size=257 * 4,
vocab_size=vocab_size,
sos=0,
eos=1,
ctc=True,
enh_type="rnn_mask_mvdr",
enh_kwargs=enh_kwargs,
asr_transform=asr_transform,
enh_transform=enh_transform,
att_type=att_type,
att_kwargs=att_kwargs,
enc_type="pytorch_rnn",
enc_proj=256,
enc_kwargs=default_rnn_enc_kwargs,
dec_dim=512,
dec_kwargs=default_rnn_dec_kwargs)
x, x_len, y, y_len, u = gen_egs(vocab_size,
batch_size,
num_channels=num_channels)
z, _, _, _ = mvdr_att_asr(x, x_len, y, y_len)
assert z.shape == th.Size([4, u + 1, vocab_size - 1])
def debug_visualize_feature():
transform = AsrTransform(feats="fbank-log-cmvn-delta",
frame_len=400,
frame_hop=160,
use_power=True,
pre_emphasis=0.97,
num_mels=80,
min_freq=20,
aug_prob=1,
norm_per_band=True,
aug_freq_args=(40, 1),
aug_time_args=(100, 1),
aug_mask_zero=False,
delta_as_channel=True)
feats, _ = transform(th.from_numpy(egs1_wav[None, ...]), None)
print(transform)
from aps.plot import plot_feature
plot_feature(feats[0, 1].numpy(), "egs")
def test_xfmr_encoder(enc_type, enc_kwargs):
nnet_cls = aps_asr_nnet("asr@xfmr")
vocab_size = 100
batch_size = 4
asr_transform = AsrTransform(feats="fbank-log-cmvn",
frame_len=400,
frame_hop=160,
window="hamm")
xfmr_encoders = ["xfmr_abs", "xfmr_rel", "xfmr_xl", "cfmr_xl"]
xfmr_asr = nnet_cls(
input_size=80,
vocab_size=vocab_size,
sos=0,
eos=1,
ctc=True,
asr_transform=asr_transform,
enc_type=enc_type,
enc_proj=512 if enc_type not in xfmr_encoders else None,
enc_kwargs=enc_kwargs,
dec_type="xfmr_abs",
dec_kwargs=default_xfmr_dec_kwargs)
x, x_len, y, y_len, u = gen_egs(vocab_size, batch_size)
z, _, _, _ = xfmr_asr(x, x_len, y, y_len)
assert z.shape == th.Size([4, u + 1, vocab_size - 1])