End-to-end text-to-speech conversion
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating speech from text. One of the systems includes one or more computers and one or more storage devices storing instructions that when executed by one or more computers cause the one or more computers to implement: a sequence-to-sequence recurrent neural network configured to: receive a sequence of characters in a particular natural language, and process the sequence of characters to generate a spectrogram of a verbal utterance of the sequence of characters in the particular natural language; and a subsystem configured to: receive the sequence of characters in the particular natural language, and provide the sequence of characters as input to the sequence-to-sequence recurrent neural network to obtain as output the spectrogram of the verbal utterance of the sequence of characters in the particular natural language.
1. A system comprising one or more computers and one or more non-transitory storage devices storing instructions that, when executed by one or more computers, cause the one or more computers to implement:
a sequence-to-sequence recurrent neural network configured to:
receive a sequence of characters in a particular natural language, and
process the sequence of characters to generate a spectrogram of a verbal utterance of the sequence of characters in the particular natural language; and
a subsystem configured to:
receive the sequence of characters in the particular natural language, and
provide the sequence of characters as input to the sequence-to-sequence recurrent neural network to obtain as output the spectrogram of the verbal utterance of the sequence of characters in the particular natural language, wherein the sequence-to-sequence recurrent neural network comprises:
an encoder neural network configured to:
receive the sequence of characters, and
process the sequence of characters to generate a respective encoded representation of each of the characters in the sequence; and
an attention-based decoder recurrent neural network configured to:
receive a sequence of decoder inputs; and
for each decoder input in the sequence:
process the decoder input and the encoded representations to generate r frames of the spectrogram, wherein r is an integer greater than one, wherein each of the second and subsequent decoder inputs in the sequence is one or more of the r frames of the spectrogram that were generated by processing the preceding decoder input in the sequence,
wherein the subsystem is further configured to generate a waveform from the spectrogram of the verbal utterance of the sequence of characters in the particular natural language.
2. The system of claim 1 , wherein the encoder neural network comprises:
an encoder pre-net neural network configured to:
receive a respective embedding of each character in the sequence, and
process the respective embedding of each character in the sequence to generate a transformed embedding of the character, and
an encoder CBHG neural network configured to:
receive the transformed embeddings, and
process the transformed embeddings to generate the encoded representations.
3. The system of claim 2 , wherein the encoder CBHG neural network comprises a bank of 1-D convolutional filters, followed by a highway network, and followed by a bidirectional recurrent neural network.
4. The system of claim 3 , wherein the bidirectional recurrent neural network is a gated recurrent unit neural network.
5. The system of claim 3 , wherein the encoder CBHG includes a residual connection between the transformed embeddings and outputs of the bank of 1-D convolutional filters.
6. The system of claim 3 , wherein the bank of 1-D convolutional filters includes a max pooling along time layer with stride one.
7. The system of claim 1 , wherein a first decoder input in the sequence is a predetermined initial frame.
8. The system of claim 1 , wherein the spectrogram is a compressed spectrogram.
9. The system of claim 8 , wherein the compressed spectrogram is a mel-scale spectrogram.
10. The system of claim 8 , wherein the system further comprises:
a post-processing neural network configured to:
receive the compressed spectrogram, and
process the compressed spectrogram to generate a waveform synthesizer input; and
a waveform synthesizer configured to:
receive the waveform synthesizer input, and
process the waveform synthesizer input to generate the waveform of the verbal utterance of the input sequence of characters in the particular natural language; and
wherein the subsystem is further configured to:
provide the compressed spectrogram as input to the post-processing neural network to obtain the waveform synthesizer input; and
provide the waveform synthesizer input as input to the waveform synthesizer to generate the waveform.
11. The system of claim 1 , wherein the subsystem is further configured to:
generate speech using the waveform, and
provide the generated speech for playback.
12. The system of claim 10 , wherein the waveform synthesizer input is a linear-scale spectrogram of the verbal utterance of the input sequence of characters in the particular natural language.
13. The system of claim 10 , wherein the waveform synthesizer is a trainable spectrogram to waveform inverter.
14. The system of claim 10 , wherein the post-processing neural network has been trained jointly with the sequence-to-sequence recurrent neural network.
15. The system of claim 10 , wherein the post-processing neural network is a CBHG neural network that comprises a 1-D convolutional subnetwork, followed by a highway network, and followed by a bidirectional recurrent neural network.
16. The system of claim 15 , wherein the bidirectional recurrent neural network is a gated recurrent unit neural network.
17. The system of claim 15 , wherein the CBHG neural network includes one or more residual connections.
18. The system of claim 15 , wherein the 1-D convolutional subnetwork comprises a bank of 1-D convolutional filters followed by a max pooling along time layer with stride one.
19. The system of claim 1 , wherein the subsystem is further configured to:
generate speech using the spectrogram of the verbal utterance of the input sequence of characters in the particular natural language; and
provide the generated speech for playback.
20. One or more non-transitory computer storage media storing instructions that, when executed by one or more computers, cause the one or more computers to implement:
a sequence-to-sequence recurrent neural network configured to:
receive a sequence of characters in a particular natural language, and
process the sequence of characters to generate a spectrogram of a verbal utterance of the sequence of characters in the particular natural language; and
a subsystem configured to:
receive the sequence of characters in the particular natural language, and
provide the sequence of characters as input to the sequence-to-sequence recurrent neural network to obtain as output the spectrogram of the verbal utterance of the sequence of characters in the particular natural language, wherein the sequence-to-sequence recurrent neural network comprises:
an encoder neural network configured to:
receive the sequence of characters, and
process the sequence of characters to generate a respective encoded representation of each of the characters in the sequence; and
an attention-based decoder recurrent neural network configured to:
receive a sequence of decoder inputs; and
for each decoder input in the sequence:
process the decoder input and the encoded representations to generate r frames of the spectrogram, wherein r is an integer greater than one, wherein each of the second and subsequent decoder inputs in the sequence is one or more of the r frames of the spectrogram that were generated by processing the preceding decoder input in the sequence,
wherein the subsystem is further configured to generate a waveform from the spectrogram of the verbal utterance of the sequence of characters in the particular natural language.