Skip to content

tts

Author: Sashi Novitasari Affiliation: NAIST (-2022) Date: 2022.08

Author: Heli Qi Affiliation: NAIST Date: 2022.09

TTSEncoder

Bases: Module

Source code in speechain/module/encoder/tts.py 
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
class TTSEncoder(Module):
    """"""

    def module_init(
        self,
        embedding: Dict,
        encoder: Dict,
        prenet: Dict = None,
        vocab_size: int = None,
    ):
        """

        Args:
            encoder:
            embedding:
            prenet:
            vocab_size:

        """
        # temporary register for connecting two sequential modules
        _prev_output_size = None

        # Token embedding layer
        embedding_class = import_class("speechain.module." + embedding["type"])
        embedding["conf"] = (
            dict() if "conf" not in embedding.keys() else embedding["conf"]
        )
        self.embedding = embedding_class(vocab_size=vocab_size, **embedding["conf"])
        _prev_output_size = self.embedding.output_size

        # TTS Encoder Prenet
        if prenet is not None:
            prenet_class = import_class("speechain.module." + prenet["type"])
            prenet["conf"] = dict() if "conf" not in prenet.keys() else prenet["conf"]
            self.prenet = prenet_class(input_size=_prev_output_size, **prenet["conf"])
            _prev_output_size = self.prenet.output_size

        # main body of the E2E TTS encoder
        encoder_class = import_class("speechain.module." + encoder["type"])
        encoder["conf"] = dict() if "conf" not in encoder.keys() else encoder["conf"]
        self.encoder = encoder_class(input_size=_prev_output_size, **encoder["conf"])

        # register the output size for assembly
        self.output_size = self.encoder.output_size

    def forward(self, text: torch.Tensor, text_len: torch.Tensor):
        """

        Args:
            text: (torch size: batch, maxlen)
            text_len: (torch size: batch )

        Returns:
            enc_outputs: dict
        """
        # Token Embedding
        text = self.embedding(text)

        # Prenet Processing if specified
        if hasattr(self, "prenet"):
            text, text_len = self.prenet(text, text_len)

        # TTS Encoding
        text, text_mask, attmat, hidden = self.encoder(
            text, make_mask_from_len(text_len)
        )
        return text, text_mask, attmat, hidden

forward(text, text_len)

Parameters:

Name Type Description Default
text Tensor

(torch size: batch, maxlen)

 required
text_len Tensor

(torch size: batch )

 required

Returns:

Name Type Description
enc_outputs

dict
Source code in speechain/module/encoder/tts.py 
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
def forward(self, text: torch.Tensor, text_len: torch.Tensor):
    """

    Args:
        text: (torch size: batch, maxlen)
        text_len: (torch size: batch )

    Returns:
        enc_outputs: dict
    """
    # Token Embedding
    text = self.embedding(text)

    # Prenet Processing if specified
    if hasattr(self, "prenet"):
        text, text_len = self.prenet(text, text_len)

    # TTS Encoding
    text, text_mask, attmat, hidden = self.encoder(
        text, make_mask_from_len(text_len)
    )
    return text, text_mask, attmat, hidden

module_init(embedding, encoder, prenet=None, vocab_size=None)

Parameters:

Name Type Description Default
encoder Dict
required
embedding Dict
required
prenet Dict
None
vocab_size int
None
Source code in speechain/module/encoder/tts.py 
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
def module_init(
    self,
    embedding: Dict,
    encoder: Dict,
    prenet: Dict = None,
    vocab_size: int = None,
):
    """

    Args:
        encoder:
        embedding:
        prenet:
        vocab_size:

    """
    # temporary register for connecting two sequential modules
    _prev_output_size = None

    # Token embedding layer
    embedding_class = import_class("speechain.module." + embedding["type"])
    embedding["conf"] = (
        dict() if "conf" not in embedding.keys() else embedding["conf"]
    )
    self.embedding = embedding_class(vocab_size=vocab_size, **embedding["conf"])
    _prev_output_size = self.embedding.output_size

    # TTS Encoder Prenet
    if prenet is not None:
        prenet_class = import_class("speechain.module." + prenet["type"])
        prenet["conf"] = dict() if "conf" not in prenet.keys() else prenet["conf"]
        self.prenet = prenet_class(input_size=_prev_output_size, **prenet["conf"])
        _prev_output_size = self.prenet.output_size

    # main body of the E2E TTS encoder
    encoder_class = import_class("speechain.module." + encoder["type"])
    encoder["conf"] = dict() if "conf" not in encoder.keys() else encoder["conf"]
    self.encoder = encoder_class(input_size=_prev_output_size, **encoder["conf"])

    # register the output size for assembly
    self.output_size = self.encoder.output_size