Processing sequences using convolutional neural networks
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing sequences using convolutional neural networks. One of the methods includes, for each of the time steps: providing a current sequence of audio data as input to a convolutional subnetwork, wherein the current sequence comprises the respective audio sample at each time step that precedes the time step in the output sequence, and wherein the convolutional subnetwork is configured to process the current sequence of audio data to generate an alternative representation for the time step; and providing the alternative representation for the time step as input to an output layer, wherein the output layer is configured to: process the alternative representation to generate an output that defines a score distribution over a plurality of possible audio samples for the time step.
1. A neural network system implemented by one or more computers,
wherein the neural network system is configured to generate an output sequence of data elements that comprises a respective data element at each of a plurality of time steps, and
wherein the neural network system comprises:
a convolutional subnetwork comprising one or more data-processing convolutional neural network layers, wherein the data-processing convolutional neural network layers are causal convolutional neural network layers, wherein the convolutional subnetwork is configured to, for each of the plurality of time steps:
receive a current sequence of data elements that comprises the respective data element at each time step that precedes the time step in the output sequence, and
process the current sequence of data elements to generate an alternative representation for the time step; and
an output layer, wherein the output layer is configured to, for each of the plurality of time steps:
receive the alternative representation for the time step, and
process the alternative representation for the time step to generate an output that defines a score distribution over a plurality of possible data elements for the time step.
2. The neural network system of claim 1 , wherein the neural network system further comprises:
a subsystem configured to, for each of the plurality of time steps:
select a data element at the time step in the output sequence in accordance with the score distribution for the time step.
3. The neural network system of claim 2 , wherein selecting the data element comprises:
sampling from the score distribution.
4. The neural network system of claim 2 , wherein selecting the data element comprises:
selecting a data element having a highest score according to the score distribution.
5. The neural network system of claim 1 , wherein each of the plurality of time steps corresponds to a respective time in an audio waveform, and wherein the respective data element at each of the plurality of time steps is an audio sample of the audio waveform at the corresponding time.
6. The neural network system of claim 5 , wherein the respective at each of the plurality of time steps is (i) a compressed or a companded representation of the audio waveform at the corresponding time, or (ii) an amplitude value of the audio waveform at the corresponding time.
7. The neural network system of claim 1 , wherein the data-processing convolutional neural network layers include one or more dilated convolutional neural network layers.
8. The neural network system of claim 7 , wherein the data-processing convolutional neural network layers include multiple blocks of dilated convolutional neural network layers, wherein each block comprises multiple dilated convolutional neural network layers with increasing dilation.
9. The neural network system of claim 1 , wherein one or more of the data-processing convolutional neural network layers have gated activation units.
10. The neural network system of claim 1 , wherein, at each of the plurality of time steps, the alternative representation is conditioned on a neural network input.
11. The neural network system of claim 10 , wherein the neural network input comprises features of a text segment, and wherein the output sequence represents a verbalization of the text segment.
12. The neural network system of claim 11 , wherein the neural network input further comprises intonation pattern values.
13. The neural network system of claim 10 , wherein the neural network input comprises one or more of: speaker identity information, language identity information, and speaking style information.
14. The neural network system of claim 1 , wherein the output sequence represents a piece of music.
15. The neural network system of claim 1 , wherein the convolutional subnetwork comprises residual connections.
16. The neural network system of claim 1 , wherein the convolutional subnetwork comprises skip connections.
17. The neural network system of claim 1 , wherein processing the current sequence of data elements to generate an alternative representation for the time step comprises reusing values computed for previous time steps.
18. One or more non-transitory computer storage media encoded with instructions that when executed by one or more computers cause the one or more computers to implement a neural network system,
wherein the neural network system is configured to generate an output sequence of data elements that comprises a respective data element at each of a plurality of time steps, and
wherein the neural network system comprises:
a convolutional subnetwork comprising one or more data-processing convolutional neural network layers, wherein the data-processing convolutional neural network layers are causal convolutional neural network layers, wherein the convolutional subnetwork is configured to, for each of the plurality of time steps:
receive a current sequence of data elements that comprises the respective data element at each time step that precedes the time step in the output sequence, and
process the current sequence of data elements to generate an alternative representation for the time step; and
an output layer, wherein the output layer is configured to, for each of the plurality of time steps:
receive the alternative representation for the time step, and
process the alternative representation for the time step to generate an output that defines a score distribution over a plurality of possible data elements for the time step.
19. A method of generating an output sequence of data elements that comprises a respective data element at each of a plurality of time steps,
wherein the method comprises, for each of the plurality of time steps:
providing a current sequence of data elements as input to a convolutional subnetwork comprising one or more data-processing convolutional neural network layers,
wherein the data-processing convolutional neural network layers are causal convolutional neural network layers,
wherein the current sequence comprises the respective data element at each time step that precedes the time step in the output sequence, and
wherein the convolutional subnetwork is configured to, for each of the plurality of time steps:
receive the current sequence of data elements, and
process the current sequence of data elements to generate an alternative representation for the time step; and
providing the alternative representation for the time step as input to an output layer, wherein the output layer is configured to, for each of the plurality of time steps:
receive the alternative representation for the time step, and
process the alternative representation for the time step to generate an output that defines a score distribution over a plurality of possible data elements for the time step.
20. The method of claim 19 , further comprising, for each of the plurality of time steps:
selecting a data element at the time step in the output sequence in accordance with the score distribution for the time step.