File size: 6,448 Bytes
12da6cc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
import haiku as hk
import jax
import jax.numpy as jnp
from jax.numpy import ndarray

from .config import FLAGS, AcousticInput, DurationInput


class TokenEncoder(hk.Module):
    """Encode phonemes/text to vector"""

    def __init__(self, vocab_size, lstm_dim, dropout_rate, is_training=True):
        super().__init__()
        self.is_training = is_training
        self.embed = hk.Embed(vocab_size, lstm_dim)
        self.conv1 = hk.Conv1D(lstm_dim, 3, padding="SAME")
        self.conv2 = hk.Conv1D(lstm_dim, 3, padding="SAME")
        self.conv3 = hk.Conv1D(lstm_dim, 3, padding="SAME")
        self.bn1 = hk.BatchNorm(True, True, 0.9)
        self.bn2 = hk.BatchNorm(True, True, 0.9)
        self.bn3 = hk.BatchNorm(True, True, 0.9)
        self.lstm_fwd = hk.LSTM(lstm_dim)
        self.lstm_bwd = hk.ResetCore(hk.LSTM(lstm_dim))
        self.dropout_rate = dropout_rate

    def __call__(self, x, lengths):
        x = self.embed(x)
        x = jax.nn.relu(self.bn1(self.conv1(x), is_training=self.is_training))
        if self.is_training:
            x = hk.dropout(hk.next_rng_key(), self.dropout_rate, x)
        x = jax.nn.relu(self.bn2(self.conv2(x), is_training=self.is_training))
        if self.is_training:
            x = hk.dropout(hk.next_rng_key(), self.dropout_rate, x)
        x = jax.nn.relu(self.bn3(self.conv3(x), is_training=self.is_training))
        if self.is_training:
            x = hk.dropout(hk.next_rng_key(), self.dropout_rate, x)
        B, L, _ = x.shape
        mask = jnp.arange(0, L)[None, :] >= (lengths[:, None] - 1)
        h0c0_fwd = self.lstm_fwd.initial_state(B)
        new_hx_fwd, _ = hk.dynamic_unroll(self.lstm_fwd, x, h0c0_fwd, time_major=False)
        x_bwd, mask_bwd = jax.tree_map(lambda x: jnp.flip(x, axis=1), (x, mask))
        h0c0_bwd = self.lstm_bwd.initial_state(B)
        new_hx_bwd, _ = hk.dynamic_unroll(
            self.lstm_bwd, (x_bwd, mask_bwd), h0c0_bwd, time_major=False
        )
        x = jnp.concatenate((new_hx_fwd, jnp.flip(new_hx_bwd, axis=1)), axis=-1)
        return x


class DurationModel(hk.Module):
    """Duration model of phonemes."""

    def __init__(self, is_training=True):
        super().__init__()
        self.is_training = is_training
        self.encoder = TokenEncoder(
            FLAGS.vocab_size,
            FLAGS.duration_lstm_dim,
            FLAGS.duration_embed_dropout_rate,
            is_training,
        )
        self.projection = hk.Sequential(
            [hk.Linear(FLAGS.duration_lstm_dim), jax.nn.gelu, hk.Linear(1)]
        )

    def __call__(self, inputs: DurationInput):
        x = self.encoder(inputs.phonemes, inputs.lengths)
        x = jnp.squeeze(self.projection(x), axis=-1)
        x = jax.nn.softplus(x)
        return x


class AcousticModel(hk.Module):
    """Predict melspectrogram from aligned phonemes"""

    def __init__(self, is_training=True):
        super().__init__()
        self.is_training = is_training
        self.encoder = TokenEncoder(
            FLAGS.vocab_size, FLAGS.acoustic_encoder_dim, 0.5, is_training
        )
        self.decoder = hk.deep_rnn_with_skip_connections(
            [hk.LSTM(FLAGS.acoustic_decoder_dim), hk.LSTM(FLAGS.acoustic_decoder_dim)]
        )
        self.projection = hk.Linear(FLAGS.mel_dim)

        # prenet
        self.prenet_fc1 = hk.Linear(256, with_bias=False)
        self.prenet_fc2 = hk.Linear(256, with_bias=False)
        # posnet
        self.postnet_convs = [hk.Conv1D(FLAGS.postnet_dim, 5) for _ in range(4)]
        self.postnet_convs.append(hk.Conv1D(FLAGS.mel_dim, 5))
        self.postnet_bns = [hk.BatchNorm(True, True, 0.9) for _ in range(4)] + [None]

    def prenet(self, x, dropout=0.5):
        x = jax.nn.relu(self.prenet_fc1(x))
        x = hk.dropout(hk.next_rng_key(), dropout, x)
        x = jax.nn.relu(self.prenet_fc2(x))
        x = hk.dropout(hk.next_rng_key(), dropout, x)
        return x

    def upsample(self, x, durations, L):
        ruler = jnp.arange(0, L)[None, :]  # B, L
        end_pos = jnp.cumsum(durations, axis=1)
        mid_pos = end_pos - durations / 2  # B, T

        d2 = jnp.square((mid_pos[:, None, :] - ruler[:, :, None])) / 10.0
        w = jax.nn.softmax(-d2, axis=-1)
        hk.set_state("attn", w[0])
        x = jnp.einsum("BLT,BTD->BLD", w, x)
        return x

    def postnet(self, mel: ndarray) -> ndarray:
        x = mel
        for conv, bn in zip(self.postnet_convs, self.postnet_bns):
            x = conv(x)
            if bn is not None:
                x = bn(x, is_training=self.is_training)
                x = jnp.tanh(x)
            x = hk.dropout(hk.next_rng_key(), 0.5, x) if self.is_training else x
        return x

    def inference(self, tokens, durations, n_frames):
        B, L = tokens.shape
        lengths = jnp.array([L], dtype=jnp.int32)
        x = self.encoder(tokens, lengths)
        x = self.upsample(x, durations, n_frames)

        def loop_fn(inputs, state):
            cond = inputs
            prev_mel, hxcx = state
            prev_mel = self.prenet(prev_mel)
            x = jnp.concatenate((cond, prev_mel), axis=-1)
            x, new_hxcx = self.decoder(x, hxcx)
            x = self.projection(x)
            return x, (x, new_hxcx)

        state = (
            jnp.zeros((B, FLAGS.mel_dim), dtype=jnp.float32),
            self.decoder.initial_state(B),
        )
        x, _ = hk.dynamic_unroll(loop_fn, x, state, time_major=False)
        residual = self.postnet(x)
        return x + residual

    def __call__(self, inputs: AcousticInput):
        x = self.encoder(inputs.phonemes, inputs.lengths)
        x = self.upsample(x, inputs.durations, inputs.mels.shape[1])
        mels = self.prenet(inputs.mels)
        x = jnp.concatenate((x, mels), axis=-1)
        B, L, _ = x.shape
        hx = self.decoder.initial_state(B)

        def zoneout_decoder(inputs, prev_state):
            x, mask = inputs
            x, state = self.decoder(x, prev_state)
            state = jax.tree_map(
                lambda m, s1, s2: s1 * m + s2 * (1 - m), mask, prev_state, state
            )
            return x, state

        mask = jax.tree_map(
            lambda x: jax.random.bernoulli(hk.next_rng_key(), 0.1, (B, L, x.shape[-1])),
            hx,
        )
        x, _ = hk.dynamic_unroll(zoneout_decoder, (x, mask), hx, time_major=False)
        x = self.projection(x)
        residual = self.postnet(x)
        return x, x + residual