Phonemes and graphemes for neural text-to-speech
A method includes receiving a text input including a sequence of words represented as an input encoder embedding. The input encoder embedding includes a plurality of tokens, with the plurality of tokens including a first set of grapheme tokens representing the text input as respective graphemes and a second set of phoneme tokens representing the text input as respective phonemes. The method also includes, for each respective phoneme token of the second set of phoneme tokens: identifying a respective word of the sequence of words corresponding to the respective phoneme token and determining a respective grapheme token representing the respective word of the sequence of words corresponding to the respective phoneme token. The method also includes generating an output encoder embedding based on a relationship between each respective phoneme token and the corresponding grapheme token determined to represent a same respective word as the respective phoneme token.
1 . A computer-implemented method executed by data processing hardware causes the data processing hardware to perform operations comprising:
receiving, at an encoder of a speech synthesis model, a text input comprising a sequence of words represented as an input encoder embedding, the input encoder embedding comprising set of grapheme tokens representing the text input as respective graphemes and a set of phoneme tokens representing the text input as respective phonemes, wherein the input encoder embedding represents, for each respective word in the sequence of words, respective sub-word level positions for both one or more of the grapheme tokens from the set of grapheme tokens that correspond to the respective word and one or more of the phoneme tokens from the set of phoneme tokens that correspond to the respective word;
for each respective phoneme token of the set of phoneme tokens:
identifying, by the encoder, a respective word of the sequence of words corresponding to the respective phoneme token; and
determining, by the encoder, one or more respective grapheme tokens representing the same respective word of the sequence of words corresponding to the respective phoneme token based on the respective sub-word level position for the respective phoneme token that corresponds to the respective word and the respective sub-word level position for each of the one or more respective grapheme tokens that correspond to the same respective word; and
generating, by the encoder, an output encoder embedding for use by a decoder of the speech synthesis model to synthesize speech from the text input, the output encoder embedding based on a relationship between each respective phoneme token and the one or more respective grapheme tokens determined to represent a same respective word as the respective phoneme token.
2 . The method of claim 1 , wherein the input encoder embedding further represents a combination of:
a segment embedding; and
a position embedding.
3 . The method of claim 2 , wherein the position embedding represents an overall index of position for each grapheme token of the set of grapheme tokens and each phoneme token of the set of phoneme tokens of the input encoder embedding.
4 . The method of claim 1 , wherein the speech synthesis model comprises an attention mechanism in communication with the encoder.
5 . The method of claim 1 , wherein the speech synthesis model comprises a duration-based upsampler in communication with the encoder.
6 . The method of claim 1 , wherein the input encoder embedding further comprises a special token identifying a language of the input text.
7 . The method of claim 1 , wherein the encoder of the speech synthesis model is pre-trained by:
feeding the encoder a plurality of training examples, each training example represented as a sequence of training grapheme tokens corresponding to a training sequence of words and a sequence of training phoneme tokens corresponding to the same training sequence of words;
masking a training phoneme token from the sequence of training phoneme tokens for a respective word from the training sequence of words; and
masking a training grapheme token from the sequence of training phoneme tokens for the respective word from the training sequence of words.
8 . The method of claim 1 , wherein:
the speech synthesis model comprises a multilingual speech synthesis model; and
the encoder of the speech synthesis model is pre-trained using a classification objective to predict a classification token of the input encoder embedding, the classification token comprising a language identifier.
9 . The method of claim 1 , wherein:
the speech synthesis model comprises a multilingual speech synthesis model; and
the output encoder embedding comprises a sequence of encoder tokens, each encoder token comprising language information about the input text.
10 . The method of claim 1 , wherein:
the speech synthesis model comprises a multi-accent speech synthesis model; and
the encoder of the speech synthesis model is pre-trained using a classification objective to predict a classification token, the classification token comprising an accent identifier.
11 . A system comprising:
data processing hardware; and
memory hardware in communication with the data processing hardware, the memory hardware storing instructions that when executed on the data processing hardware cause the data processing hardware to perform operations comprising:
receiving, at an encoder of a speech synthesis model, a text input comprising a sequence of words represented as an input encoder embedding, the input encoder embedding comprising a set of grapheme tokens representing the text input as respective graphemes and a set of phoneme tokens representing the text input as respective phonemes, wherein the input encoder embedding represents, for each respective word in the sequence of words, respective sub-word level positions for both one or more of the grapheme tokens from the set of grapheme tokens that correspond to the respective word and one or more of the phoneme tokens from the set of phoneme tokens that correspond to the respective word;
for each respective phoneme token of the set of phoneme tokens:
identifying, by the encoder, a respective word of the sequence of words corresponding to the respective phoneme token; and
determining, by the encoder, one or more respective grapheme tokens representing the same respective word of the sequence of words corresponding to the respective phoneme token based on the respective sub-word level position for the respective phoneme token that corresponds to the respective word and the respective sub-word level position for each of the one or more respective grapheme tokens that correspond to the same respective word; and
generating, by the encoder, an output encoder embedding for use by a decoder of the speech synthesis model to synthesize speech from the text input, the output encoder embedding based on a relationship between each respective phoneme token and the one or more respective grapheme tokens determined to represent a same respective word as the respective phoneme token.
12 . The system of claim 11 , wherein the input encoder embedding further represents a combination of:
a segment embedding; and
a position embedding.
13 . The system of claim 12 , wherein the position embedding represents an overall index of position for each grapheme token of the set of grapheme tokens and each phoneme token of the set of phoneme tokens of the input encoder embedding.
14 . The system of claim 11 , wherein the speech synthesis model comprises an attention mechanism in communication with the encoder.
15 . The system of claim 11 , wherein the speech synthesis model comprises a duration-based upsampler in communication with the encoder.
16 . The system of claim 11 , wherein the input encoder embedding further comprises a special token identifying a language of the input text.
17 . The system of claim 11 , wherein the encoder of the speech synthesis model is pre-trained by:
feeding the encoder a plurality of training examples, each training example represented as a sequence of training grapheme tokens corresponding to a training sequence of words and a sequence of training phoneme tokens corresponding to the same training sequence of words;
masking a training phoneme token from the sequence of training phoneme tokens for a respective word from the training sequence of words; and
masking a training grapheme token from the sequence of training phoneme tokens for the respective word from the training sequence of words.
18 . The system of claim 11 , wherein:
the speech synthesis model comprises a multilingual speech synthesis model; and
the encoder of the speech synthesis model is pre-trained using a classification objective to predict a classification token of the input encoder embedding, the classification token comprising a language identifier.
19 . The system of claim 11 , wherein:
the speech synthesis model comprises a multilingual speech synthesis model; and
the output encoder embedding comprises a sequence of encoder tokens, each encoder token comprising language information about the input text.
20 . The system of claim 11 , wherein:
the speech synthesis model comprises a multi-accent speech synthesis model; and
the encoder of the speech synthesis model is pre-trained using a classification objective to predict a classification token, the classification token comprising an accent identifier.