Generating audio using auto-regressive generative neural networks
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating a prediction of an audio signal. One of the methods includes receiving a request to generate an audio signal; obtaining a semantic representation of the audio signal; generating, using one or more generative neural networks and conditioned on at least the semantic representation, an acoustic representation of the audio signal; and processing at least the acoustic representation using a decoder neural network to generate the prediction of the audio signal.
1 . A computer-implemented method for generating an acoustic representation of an audio signal, the method comprising:
receiving a request to generate an audio signal having a respective audio sample at each of a plurality of output time steps spanning a time window;
obtaining a semantic representation of the audio signal that specifies a respective semantic token at each of a plurality of first time steps spanning the time window; and
generating, using one or more generative neural networks and conditioned on at least the semantic representation, the acoustic representation of the audio signal, the acoustic representation specifying a set of one or more respective acoustic tokens at each of a plurality of second time steps spanning the time window, wherein the set of one or more respective acoustic tokens at each of the plurality of second time steps comprises a plurality of acoustic tokens that collectively represent a prediction of an output of a residual vector quantization applied to an embedding that represents acoustic properties of the audio signal at the second time step,
the residual vector quantization encodes the embedding using a hierarchy of a plurality of vector quantizers that each generate a respective acoustic token from a corresponding vocabulary of acoustic tokens for the vector quantizer, wherein the hierarchy comprises one or more coarse vector quantizers and one or more fine vector quantizers, and
the set of acoustic tokens at each second time step comprises, for each vector quantizer, a respective acoustic token selected from the vocabulary for the vector quantizer.
2 . The method of claim 1 , wherein each semantic token is selected from a vocabulary of semantic tokens and represents semantic content of the audio signal at the corresponding first time step.
3 . The method of claim 1 , wherein the one or more respective acoustic tokens at each second time step represent acoustic properties of the audio signal at the corresponding second time step.
4 . The method of claim 1 , further comprising:
processing at least the acoustic representation using a decoder neural network to generate a prediction of the audio signal.
5 . The method of claim 4 , wherein the decoder neural network is a decoder neural network of a neural audio codec that has been trained jointly with an encoder neural network on an objective that measures reconstruction quality of predicted audio signals generated by the decoder neural network from acoustic representations generated using outputs generated by the encoder neural network.
6 . The method of claim 1 , wherein the acoustic representation is a prediction of a ground truth acoustic representation that would be generated from outputs of an encoder neural network by processing the audio signal.
7 . The method of claim 6 , wherein the encoder neural network outputs a respective embedding at each of the plurality of second time steps, and wherein the ground truth acoustic representation is generated by applying quantization to each of the respective embeddings.
8 . The method of claim 7 , wherein:
the quantization is residual vector quantization that encodes each respective embedding using the hierarchy of the plurality of vector quantizers, and
the set of one or more respective acoustic tokens at each second time step comprise, for each vector quantizer, a respective acoustic token that is a prediction of a ground truth acoustic token that would be generated by the vector quantizer from a ground truth embedding generated by the encoder neural network at the second time step.
9 . The method of claim 1 , wherein the hierarchy comprises the one or more coarse vector quantizers at one or more first positions in the hierarchy and the one or more fine vector quantizers at one or more last positions in the hierarchy.
10 . The method of claim 1 , wherein generating, using one or more generative neural networks and conditioned on at least the semantic representation, an acoustic representation of the audio signal comprises:
generating, using a first generative neural network and for each of the one or more coarse vector quantizers in the hierarchy, the respective acoustic tokens for the second time steps for the coarse vector quantizer conditioned on at least the semantic representation.
11 . The method of claim 10 , wherein the first generative neural network is an auto-regressive neural network that is configured to generate the acoustic tokens auto-regressively according to a first generation order, and wherein each particular acoustic token for each particular coarse vector quantizer and at each particular second time step is conditioned on at least the semantic representation and any acoustic tokens that precede the particular acoustic token in the first generation order.
12 . The method of claim 11 , wherein each particular acoustic token for each particular coarse vector quantizer and at each particular second time step is preceded in the first generation order by (i) any acoustic token for any of the coarse vector quantizers at any second time step that precedes the particular second time step and (ii) any acoustic tokens at the particular second time step for any coarse vector quantizers that precede the particular coarse vector quantizer in the hierarchy.
13 . The method of claim 10 , wherein the first generative neural network has a decoder-only Transformer architecture or an encoder-decoder Transformer architecture.
14 . The method of claim 10 , wherein generating, using one or more generative neural networks and conditioned on at least the semantic representation, an acoustic representation of the audio signal comprises:
generating, using a second generative neural network and for each of the one or more fine vector quantizers in the hierarchy, the respective acoustic tokens for the second time steps for the fine vector quantizer conditioned on the respective acoustic tokens for the second time steps for the one or more coarse vector quantizers in the hierarchy.
15 . The method of claim 14 , wherein the second generative neural network is not conditioned on the semantic representation.
16 . The method of claim 14 , wherein the second generative neural network is an auto-regressive neural network that is configured to generate the acoustic tokens auto-regressively according to a second generation order, and wherein each particular acoustic token for each particular fine vector quantizer and at each particular second time step is conditioned on (i) the respective acoustic tokens for at least a subset of the second time steps for the one or more coarse vector quantizers and (ii) at least a subset of the acoustic tokens that precede the particular acoustic token in the second generation order.
17 . The method of claim 16 , wherein each particular acoustic token for each particular fine vector quantizer and at each particular second time step is preceded in the second generation order by (i) any acoustic token for any of the fine vector quantizers at any second time step that precedes the particular second time step and (ii) any acoustic tokens at the particular second time step for any fine vector quantizers that precede the particular fine vector quantizer in the hierarchy.
18 . The method of claim 16 , wherein each particular acoustic token for each particular fine vector quantizer and at each particular second time step is conditioned on (i) the respective acoustic tokens for the one or more coarse vector quantizers that are at most a threshold number of second time steps before the second time step and (ii) any acoustic tokens that precede the particular second time step in the second generation order and that are at second time steps that are at most a threshold number of second time steps before the second time step.
19 . The method of claim 14 , wherein the second generative neural network has a decoder-only Transformer architecture or an encoder-decoder Transformer architecture.
20 . The method of claim 1 , wherein obtaining a semantic representation of the audio signal comprises:
generating the semantic representation auto-regressively using a third generative neural network.
21 . The method of claim 1 , wherein the request specifies a context for the audio signal and the audio signal is conditioned on the context.
22 . The method of claim 21 , wherein the context specifies semantic properties of the audio signal and wherein obtaining a semantic representation of the audio signal comprises:
generating the semantic representation conditioned on the context.
23 . The method of claim 21 , wherein the context specifies acoustic properties of the audio signal, and wherein generating, using one or more generative neural networks and conditioned on at least the semantic representation, the acoustic representation of the audio signal comprises:
generating, using one or more generative neural networks and conditioned on the semantic representation and the context, the acoustic representation of the audio signal.
24 . The method of claim 23 , wherein the set of one or more respective acoustic tokens at each of the plurality of second time steps comprises a plurality of acoustic tokens that collectively represent a prediction of an output of a residual vector quantization applied to an embedding that represents acoustic properties of the audio signal at the second time step,
the residual vector quantization encodes the embedding using a hierarchy of a plurality of vector quantizers that each generate a respective acoustic token from a corresponding vocabulary of acoustic tokens for the vector quantizer, wherein the hierarchy comprises one or more coarse vector quantizers at one or more first positions in the hierarchy and one or more fine vector quantizers at one or more last positions in the hierarchy, and
the set of acoustic tokens at each second time step comprises, for each vector quantizer, a respective acoustic token selected from the vocabulary for the vector quantizer, and wherein generating, using one or more generative neural networks and conditioned on the semantic representation and the context, an acoustic representation of the audio signal comprises:
generating, using a first generative neural network and for each of the one or more coarse vector quantizers in the hierarchy, the respective acoustic tokens for the second time steps for the coarse vector quantizer conditioned on the semantic representation and the context.
25 . The method of claim 23 , wherein processing at least the acoustic representation using a decoder neural network to generate the prediction of the audio signal comprises:
processing the acoustic representation and an acoustic representation of the context using the decoder neural network to generate the prediction of the audio signal.
26 . The method of claim 21 , wherein the context comprises an audio input.
27 . The method of claim 21 , wherein the context comprises visual data.
28 . The method of claim 21 , wherein the context comprises text data.
29 . The method of claim 1 , wherein a number of first time steps and a number of second time steps that span the time window is less than the number of output time steps that span the time window.
30 . The method of claim 29 , wherein the number of first time steps that span the time window is less than the number of second time steps that span the time window.
31 . A system comprising one or more computers and one or more storage devices storing instructions that, when executed by the one or more computers, cause the one or more computers to perform operations for generating an acoustic representation of an audio signal, wherein the operations comprise:
receiving a request to generate an audio signal having a respective audio sample at each of a plurality of output time steps spanning a time window;
obtaining a semantic representation of the audio signal that specifies a respective semantic token at each of a plurality of first time steps spanning the time window; and
generating, using one or more generative neural networks and conditioned on at least the semantic representation, the acoustic representation of the audio signal, the acoustic representation specifying a set of one or more respective acoustic tokens at each of a plurality of second time steps spanning the time window, wherein the set of one or more respective acoustic tokens at each of the plurality of second time steps comprises a plurality of acoustic tokens that collectively represent a prediction of an output of a residual vector quantization applied to an embedding that represents acoustic properties of the audio signal at the second time step,
the residual vector quantization encodes the embedding using a hierarchy of a plurality of vector quantizers that each generate a respective acoustic token from a corresponding vocabulary of acoustic tokens for the vector quantizer, wherein the hierarchy comprises one or more coarse vector quantizers and one or more fine vector quantizers, and
the set of acoustic tokens at each second time step comprises, for each vector quantizer, a respective acoustic token selected from the vocabulary for the vector quantizer.
32 . One or more non-transitory computer-readable storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations for generating an acoustic representation of an audio signal, wherein the operations comprise:
receiving a request to generate an audio signal having a respective audio sample at each of a plurality of output time steps spanning a time window;
obtaining a semantic representation of the audio signal that specifies a respective semantic token at each of a plurality of first time steps spanning the time window; and
generating, using one or more generative neural networks and conditioned on at least the semantic representation, the acoustic representation of the audio signal, the acoustic representation specifying a set of one or more respective acoustic tokens at each of a plurality of second time steps spanning the time window, wherein the set of one or more respective acoustic tokens at each of the plurality of second time steps comprises a plurality of acoustic tokens that collectively represent a prediction of an output of a residual vector quantization applied to an embedding that represents acoustic properties of the audio signal at the second time step,
the residual vector quantization encodes the embedding using a hierarchy of a plurality of vector quantizers that each generate a respective acoustic token from a corresponding vocabulary of acoustic tokens for the vector quantizer, wherein the hierarchy comprises one or more coarse vector quantizers and one or more fine vector quantizers, and
the set of acoustic tokens at each second time step comprises, for each vector quantizer, a respective acoustic token selected from the vocabulary for the vector quantizer.