WO2018048934A1 - Generating audio using neural networks - Google Patents
Generating audio using neural networks Download PDFInfo
- Publication number
- WO2018048934A1 WO2018048934A1 PCT/US2017/050320 US2017050320W WO2018048934A1 WO 2018048934 A1 WO2018048934 A1 WO 2018048934A1 US 2017050320 W US2017050320 W US 2017050320W WO 2018048934 A1 WO2018048934 A1 WO 2018048934A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- neural network
- audio
- time
- convolutional
- time step
- Prior art date
Links
- 238000013528 artificial neural network Methods 0.000 title claims description 92
- 238000000034 method Methods 0.000 claims abstract description 49
- 238000009826 distribution Methods 0.000 claims abstract description 26
- 230000008569 process Effects 0.000 claims abstract description 25
- 238000003860 storage Methods 0.000 claims abstract description 9
- 230000001364 causal effect Effects 0.000 claims description 54
- 238000012545 processing Methods 0.000 claims description 39
- 238000013527 convolutional neural network Methods 0.000 claims description 33
- 230000004913 activation Effects 0.000 claims description 26
- 230000010339 dilation Effects 0.000 claims description 22
- 230000001143 conditioned effect Effects 0.000 claims description 10
- 238000005070 sampling Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 abstract description 14
- 230000006870 function Effects 0.000 description 32
- 230000003750 conditioning effect Effects 0.000 description 11
- 238000012549 training Methods 0.000 description 7
- 238000012800 visualization Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 238000012886 linear function Methods 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 230000000306 recurrent effect Effects 0.000 description 4
- 230000003993 interaction Effects 0.000 description 3
- 230000015654 memory Effects 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000013515 script Methods 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 238000003786 synthesis reaction Methods 0.000 description 2
- 241000009334 Singa Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000010420 art technique Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 230000009850 completed effect Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000001953 sensory effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L13/00—Speech synthesis; Text to speech systems
- G10L13/06—Elementary speech units used in speech synthesisers; Concatenation rules
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10H—ELECTROPHONIC MUSICAL INSTRUMENTS; INSTRUMENTS IN WHICH THE TONES ARE GENERATED BY ELECTROMECHANICAL MEANS OR ELECTRONIC GENERATORS, OR IN WHICH THE TONES ARE SYNTHESISED FROM A DATA STORE
- G10H2250/00—Aspects of algorithms or signal processing methods without intrinsic musical character, yet specifically adapted for or used in electrophonic musical processing
- G10H2250/311—Neural networks for electrophonic musical instruments or musical processing, e.g. for musical recognition or control, automatic composition or improvisation
Definitions
- This specification relates to processing and generating audio using neural networks.
- Neural networks are machine learning models that employ one or more layers of nonlinear units to predict an output for a received input.
- Some neural networks include one or more hidden layers in addition to an output layer. The output of each hidden layer is used as input to the next layer in the network, i.e., the next hidden layer or the output layer.
- Each layer of the network generates an output from a received input in accordance with current values of a respective set of parameters.
- This specification describes how a system implemented as computer programs on one or more computers in one or more locations can generate a sequence of audio data that includes a respective audio sample at each of multiple time steps.
- the sequence of audio data can represent speech in a particular natural language or a piece of music.
- a neural network system implemented by one or more computers is configured to generate an output sequence of audio data that comprises a respective audio sample at each of a plurality of time steps.
- the neural network system may comprise a convolutional subnetwork comprising one or more audio-processing convolutional neural network layers; and an output layer.
- the convolutional subnetwork may be configured to, for each of the plurality of time steps: receive a current sequence of audio data that comprises the respective audio sample at each time step that precedes the (current) time step in the output sequence.
- the convolutional subnetwork may further be configured to process the current sequence of audio data to generate an alternative representation for the time (current) step.
- This alternative representation may thus comprise a numeric representation, i.e., an ordered collection of numeric values, in which the current sequence of audio data has been encoded by the convolutional subnetwork, for example encoding features of the current sequence.
- the output layer may be 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 audio samples for the time step.
- the system can use the score distribution to select a sample for the current time step, by sampling from the distribution.
- the output may, but need not necessarily, comprise one score for each possible audio sample value, for example 256 scores for 256 possible values. In can thus be useful to compress or compand the audio sample values, which may be amplitude values, to reduce the number of model outputs.
- the convolutional neural network layers are causal convolutional neural network layers, as described in more detail later.
- the audio-processing convolutional neural network layers may include one or more dilated causal convolutional neural network layers.
- a dilated convolutional neural network layer applies a convolution to non-adjacent values in a sequence, i.e., as defined by the outputs from a previous layer. This can increase the receptive field of the convolutional subnetwork by orders of magnitude whilst preserving the input (time) resolution and maintaining computational efficiency.
- the convolutional neural network layers include multiple stacked blocks of dilated convolutional neural network layers.
- Each block may comprise multiple dilated convolutional neural network layers with increasing dilation. For example the dilation may be increased by a factor n for each successive layer up to a limit within each block. This can further increase the receptive field size.
- one or more of the convolutional neural network layers may have gated activation units.
- a rectified linear or other unit following a convolution implemented by a layer may be replaced by a gated activation unit.
- the output may be a combination of two (causal) convolutions, a main convolution and a gate convolution.
- the convolutions may each be applied to some or all of the same outputs from the previous layer.
- the combination may involve a non-linear activation function applied to the gate convolution, for example an activation with a (0,1) range such as a sigmoid. This may then multiply a value from the main convolution; a non-linear activation function may, but need not be, applied to the main convolution.
- Such an approach may assist in capturing more complex structure within the data.
- the alternative representation from the convolutional subnetwork at each time step may be conditioned on a neural network input, for example a latent representation of a conditioning input.
- the conditioning input may be global (substantially time- independent) and/or local (time-dependent).
- the conditioning input may comprise, for example, text, image or video data, or audio data, for example an example of a particular speaker or language or music.
- the neural network input may comprise an embedding of the conditioning input.
- a global conditioning input may comprise a speaker embedding and a local conditioning input may comprise linguistic features.
- the system may be configured to map the neural network input, or a conditioning input, from a lower sampling frequency to the audio sample generation frequency, for example by repeating the input or upsampling the input using a neural network.
- the neural network input may comprise features of a text segment and the output sequence may represent a verbalization of the text segment; and/or the neural network input may comprise speaker or intonation pattern values; and/or the neural network input may include one or more of: speaker identity information, language identity information, and speaking style information.
- the output sequence represents a piece of music.
- the convolutional subnetwork may comprise residual connections, for example a connection from an input of a convolutional layer to a summer to sum this with an intermediate output of the layer. This effectively allows the network to be trained to skip or partially skip a layer, thus speeding up convergence and facilitating training of deeper models.
- the convolutional subnetwork may additionally or alternatively comprise skip connections, for example directly from each of one or more intermediate layers of the convolutional subnetwork to one or more operations that directly generate the alternative representation that is provided to the output layer.
- processing the current sequence of audio data using the convolutional subnetwork to generate an alternative representation for the time step, re-uses values computed for previous time steps.
- the re-used values may comprise values derived from application of a convolutional filter to the audio sample data or data derived therefrom.
- the re-used values may be stored at one time step and retrieved at a later time step when the same filter is applied to the same (or some of the same) audio sample data or data derived therefrom. This can make the system computationally more efficient and hence faster, because there is no need to re-compute the stored values.
- the neural network system can generate on the order of tens of thousands of audio samples per second, providing a greater level of granularity than other neural network-based audio generation systems.
- the neural network system can achieve results that significantly outperform the state of the art on audio generation tasks, e.g., by generating speech from text that is of higher quality than state of the art techniques.
- a single trained neural network system can be used to generate different voices by conditioning on the speaker identity.
- convolutional neural network layers e.g., causal convolutional layers
- recurrent neural network layers e.g., instead of long short-term memory (LSTM) layers
- LSTM long short-term memory
- the neural network system can achieve these advantageous results while not needing as many computational resources to train as other systems that do include recurrent neural network layers, resulting in a reduced training time.
- convolutional layers rather than recurrent layers, the computation of the neural network system can be more easily batched and more easily parallelized, e.g.., because the layers of the network do not have to be unrolled for each time step, allowing the computation of the system to be performed more efficiently.
- dilated causal convolutional layers the receptive field of the convolutional subnetwork and, therefore, the quality of the audio generated by the system, can be improved without greatly increasing the computational cost of generating the audio.
- FIG. 1 shows an example neural network system.
- FIG. 2 shows a visualization of an example block of dilated causal convolutional layers.
- FIG. 3 shows an example architecture for the convolutional subnetwork.
- FIG. 4 is a flow diagram of an example process for generating an audio sample at a given time step in an audio sequence.
- FIG. 1 shows an example neural network system 100.
- the neural network system 100 is an example of a system implemented as computer programs on one or more computers in one or more locations, in which the systems, components, and techniques described below can be implemented.
- the neural network system 100 generates sequences of audio data that each include a respective audio sample at each of multiple time steps, e.g., an output sequence of audio data 152.
- each time step in a given audio sequence corresponds to a respective time in an audio waveform and the audio sample at the time step characterizes the waveform at the corresponding time.
- the audio sample at each time step in the sequence is the amplitude of the audio waveform at the corresponding time, i.e., the sequence generated by the neural network system 100 is a raw audio waveform.
- the audio sample at each time step in the sequence is a compressed or companded representation of the waveform at the corresponding time.
- the audio sample can be a ⁇ -law transformed representation of the waveform.
- the neural network system 100 generates audio sequences autoregressively. That is, for each particular time step in an output audio sequence, the neural network system 100 generates the audio sample at the time step conditioned on the audio samples that have already been generated as of the particular time step, i.e., on audio samples at time steps that are earlier than the particular time step in the audio sequence.
- the neural network system 100 includes a convolutional subnetwork 110 and an output layer 120.
- the convolutional subnetwork 1 10 is configured to receive the current audio sequence, i.e., the audio sequence that has already been generated as of the time step, and to process the current audio sequence to generate an alternative representation for the time step.
- the convolutional subnetwork 1 10 can receive a current audio sequence 142 that includes the audio samples that precede the audio sample 140 in the audio sequence 152 and process the current audio sequence 142 to generate an alternative representation 144.
- the output layer 120 is configured to, at each of the time steps, receive the alternative representation at the time step and generate a score distribution over possible audio samples for the time step.
- the score distribution includes a respective score for each of multiple possible audio samples.
- the output layer 120 is a softmax output layer.
- the output layer 120 can receive the alternative representation 144 and process the alternative representation 144 to generate a score distribution 146.
- the score distribution when the neural network system 100 is configured to generate raw audio data, the score distribution includes a respective score for each of multiple possible amplitude values.
- the score distribution when the neural network system 100 is configured to generate compressed or companded values, the score distribution includes a respective score for each of multiple possible compressed or companded values.
- the neural network system 100 can select an audio sample to be included in the output sequence at the given time step from the multiple possible audio samples in accordance with the score distribution for the given time step. For example, the neural network system 100 can select an audio sample by sampling from the score distribution, i.e., sampling from the possible audio samples in accordance with the scores in the score distribution so that each audio sample is selected with a likelihood that corresponds to the score for the audio sample, or can select the possible audio sample having the highest score according to the score distribution.
- the convolutional subnetwork 110 generally includes multiple audio-processing convolutional neural network layers. More specifically, the audio-processing convolutional neural network layers include multiple causal convolutional layers.
- a causal convolutional layer is a convolutional layer that operates on an input sequence that has a respective input at each of multiple time steps by, for each time step, generating an output that depends only on the inputs at the time step and at the time steps before the time step in the input sequence, i.e., and not on any inputs at any time steps after the time step in the input sequence.
- the causal convolutional layers are implemented by applying a normal convolution and then shifting each output of the normal convolution by a few time steps, i.e., shifting each output forward by (filter length - 1) time steps, prior to applying the activation function for the convolutional layer, where "filter length" is the length of the filter of the convolution that is being applied.
- some or all of the audio-processing convolutional layers can be dilated causal convolutional layers.
- a dilated convolution is a convolution where the filter is applied over an area larger than its length by skipping input values with a certain step that is defined by the dilation value for the dilated convolution.
- the audio-processing neural network layers include a stack of multiple blocks of dilated causal convolutional layers.
- Each block in the stack can include multiple dilated convolutional neural network layers with increasing dilation.
- the dilation can double for each layer starting from an initial dilation, and then return to the initial dilation for the first layer in the next block.
- the dilations of the dilated convolutional layers in a block can be, in order: 1 , 2, 4, . . . , 512.
- a simplified example of a block of dilated causal convolutional layers is described below with reference to FIG. 2.
- the convolutional subnetwork includes residual connections, skip connections, or both.
- An example architecture of the convolutional subnetwork that includes both residual connections and skip connections is described below with reference to FIG. 3.
- the neural network system 100 generates audio sequences conditioned on a neural network input.
- the neural network system 100 can generate the audio sequence 152 conditioned on a neural network input 102.
- the neural network input includes one or more local features, i.e., one or more features that are different for different time steps in the output sequence.
- the neural network system 100 can obtain as input linguistic features of a text segment and can generate an audio sequence that represents a verbalization of the text segment, i.e., the neural network system 100 can function as part of a text-to-speech system that converts written text to spoken speech and also includes a component that verbalizes the audio sequence generated by the neural network system 100.
- the neural network input includes one or more global features, i.e., one or more features that are the same throughout the entire output sequence.
- the neural network system 100 can generate speech conditioned on an identity of the speaker, i.e., so that the speech is generated to sound like the voice of the speaker.
- the neural network system 100 can obtain a vector encoding the identity of the speaker, e.g., a one-hot encoded vector identifying the speaker, and condition the generated speech on the obtained vector.
- the audio sequences are conditioned on the neural network input by conditioning the activation function of some or all of the convolutional layers in the convolutional subnetwork. That is, the output of the activation function and, accordingly, the output of the convolutional layer, is dependent not only on the output of the convolution performed by the layer but also on the neural network input.
- FIG. 2 shows a visualization 200 of an example block of dilated causal convolutional layers.
- the example block includes a dilated causal convolutional layer 204 with dilation one, a dilated causal convolutional layer 206 with dilation two, a dilated causal convolutional layer 208 with dilation four, and a dilated causal convolutional layer 210 with dilation eight.
- the block of dilated causal convolutional layers are operating on a current input sequence 202 to generate an output sequence.
- the visualization 200 visualizes using bold arrows how the block generates the output 212 that is the output at the time step that is currently the last time step in the current input sequence 202 and the output sequence.
- each layer in the block is a causal convolutional layer
- the output 212 depends only on outputs that are at the last current time step or time steps before the last current time step in the various sequences operated on by the layers in the block.
- the layers in the block are arranged in order of increasing dilation, with the first layer in the block, i.e., dilated causal convolutional layer 204, having dilation one and the last layer in the block, i.e., dilated causal convolutional layer 204, having dilation eight.
- the filter of the layer 204 is applied to adjacent inputs in the current input sequence 202.
- the filter of the layer 206 is applied to outputs that are separated by one output in the output sequence generated by the layer 204.
- the filter of the layer 208 is applied to outputs that are separated by three outputs in the output sequence generated by the layer 206. Because the dilated causal convolutional layer 210 has dilation eight, the filter of the layer 210 is applied to outputs that are separated by seven outputs in the output sequence generated by the layer 208.
- FIG. 3 shows an example architecture 300 for the convolutional subnetwork 110 of FIG. 1.
- the dilated causal convolutional layers that are in the convolutional subnetwork have residual connections and skip connections.
- the convolutional subnetwork 1 10 includes a causal convolutional layer 302 that processes the current output sequence 142, i.e., by applying a causal convolution to the current output sequence 142.
- the convolutional subnetwork 110 then processes the output of the causal convolutional layer 302 through a stack of dilated causal convolutional layers.
- Each dilated causal convolutional layer 304 in the stack applies a dilated causal convolution 308 to the input 306 to the dilated causal convolutional layer 304.
- the dilated causal convolutional layers in the stack are arranged in blocks, with the dilation of the dilated causal convolutions applied by each layer increasing within a given block and then restarting at the initial value for the first layer in the next block.
- the dilated causal convolutional layers in the stack have a gated activation function in which the output of an element-wise non-linearity, i.e., of a conventional activation function, is element-wise multiplied by a gate vector.
- the dilated causal convolution 308 includes two dilated causal convolutions on the layer input 302 - a first dilated causal convolution between a main filter for the layer 304 and the layer input 306 and another dilated causal convolution between a gate filter for the layer 304 and the layer input 306.
- dilated causal convolution 308 is a single dilated causal convolution and half of the output of the single convolution is provided as the output of the dilated causal convolution between the main filter for the layer 304 and the layer input 306 and the other half of the output of the single convolution is provided as the output of the dilated causal convolution between the gate filter for the layer 304 and the layer input 306.
- the dilated causal convolutional layer 304 determines the output of the activation function of the layer 304 using the outputs of the dilated causal convolution.
- the layer 304 applies an element-wise non-linear function 310 which, in the example of FIG. 3 is the tank function, to the output of the dilated convolution with the main filter and applies an element- wise gating function which, in the example of FIG. 3, is the sigmoid function, to the output of the dilated convolution with the gate filter.
- the layer 304 then performs an element- wise multiplication 314 between the output of the non-linear function 310 and the output of the gating function 312 to generate the activation function output.
- the output of the activation function z for a layer satisfies:
- fc is the main filter for the layer k
- x is the layer input
- * denotes a causal dilated convolution
- ⁇ denotes element-wise multiplication
- W g k is the gate filter for the layer k.
- the layer 304 When the output sequence being generated is conditioned on a neural network input, the layer 304 also conditions the output of the activation function on the neural network input.
- the non-linear function and the gating function each take as input a combination of the corresponding dilated convolution output and an input generated from the neural network input.
- the element-wise non-linearity is tanh and the element-wise gating function is the sigmoid function
- the output of the activation function z for the layer k satisfies:
- V k is a main leamable linear projection (of h to the main component of the activation function) for the layer k
- h is the neural network input
- V k is a gate leamable linear projection (of h to the gate component of the activation function) for the layer k.
- the system 100 obtains a sequence y that includes a set of features for each time step in the output sequence.
- V j k * y and V g k * y are respective l x l convolutions.
- the system 100 directly receives the sequence y as the neural network input, i.e., directly receives a sequence that has the same resolution as the output sequence.
- the system 100 receives a sequence having a lower resolution, i.e., with a lower sampling frequency, than the output sequence.
- the system can generate the sequence y by processing the lower resolution sequence using a transposed (learned upsampling) convolutional network to generate the sequence y or can repeat values from the lower resolution sequence across time to generate the sequence y.
- the local features are linguistic features for use in text to speech generation
- the linguistic features can include some or all of phone, syllable, word, phrase, and utterance-level features of the text.
- Example sets of linguistic features that can be used are described in Zen, Heiga. An example of context-dependent label formator HMM-based speech synthesis in English, 2006. URL
- the layer 304 then performs a 1 x 1 convolution 316 on the activation function output.
- the layer 304 provides the output of the 1 x 1 convolution as the skip output 318 of the layer and adds the residual, i.e., the layer input 306, and the output of the 1 x 1 convolution to generate the final output 320 of the layer 304.
- the convolutional subnetwork 110 then provides the final output 320 as the layer input to the next dilated convolutional layer in the stack.
- the layer 304 performs two l x l convolutions on the activation function output, one with a residual filter and the other with a skip filter. In these implementations, the layer 304 provides the output of the convolution with the skip filter as the skip output 318 of the layer and adds the residual and the output of the 1 x 1 convolution with the residual filter to generate the final output 320 of the layer 304.
- the convolutional subnetwork 110 then provides the final output 320 as the layer input to the next dilated convolutional layer in the stack. For the last layer in the stack, because there is no next layer, the convolutional subnetwork 110 can either discard the final output 320 generated by the last layer or can refrain from computing a final output, i.e., can refrain from performing the 1 x 1 convolution and the residual sum for the last layer in the stack.
- the convolutional subnetwork 110 sums 322 the skip outputs generated by the layers 304.
- the convolutional subnetwork 110 can then apply one or more non-linear functions, one or more l x l convolutions, or both to the sum 322 to generate the alternative representation 144.
- the convolutional subnetwork 110 applies an element- wise non-linearity 324, e.g., a ReLU, followed by a lxl convolution 326, followed by another element-wise non-linearity 328, and followed by a final lxl convolution 330, to generate the alternative representation 144.
- the output layer 120 then processes the alternative representation 144 to generate the score distribution 146.
- FIG. 4 is a flow diagram of an example process 400 for generating an audio sample at a given time step in an audio sequence.
- the process 400 will be described as being performed by a system of one or more computers located in one or more locations.
- a neural network system e.g., the neural network system 100 of FIG.1 , appropriately programmed, can perform the process 400.
- the system provides a current audio sequence as input to the convolutional subnetwork (step 402).
- the current audio sequence is the audio sequence that has already been generated as of the given time step, i.e., a sequence that includes the output audio samples at time steps before the given time step.
- the convolutional subnetwork includes audio-processing convolutional neural network layers, e.g., dilated causal convolutional layers, and is configured to process the current sequence of audio data to generate an alternative representation for the given time step.
- the system provides the alternative representation as input to an output layer, e.g., a softmax output layer (step 404).
- the output layer is configured to process the alternative representation to generate a score distribution over possible audio samples for the time step.
- the system selects an audio sample for inclusion in the audio sequence at the given time step in accordance with the score distribution (step 406).
- the system can sample a possible audio sample in accordance with the score distribution.
- the system may be trained on raw or compressed and/or companded audio data, for example waveforms of human speakers, music and so forth.
- Optionally conditioning data may be included, for example text-to-speech data, which may be represented as linguistic features derived from text paired with audio data for a verbalization of the text.
- the system can generate the conditional predictions for all time steps in parallel, i.e., instead of autoregressively, because all time steps of the ground truth output that should be generated by the system are known.
- the system can perform the training to determine the trained values of the parameters using conventional supervised learning techniques, e.g., a stochastic gradient descent with backpropagation based technique.
- supervised learning techniques e.g., a stochastic gradient descent with backpropagation based technique.
- the system does not need as many computational resources to train as other systems, e.g., those that include recurrent neural network layers, resulting in a reduced training time.
- the convolutional subnetwork reuses values computed for previous time steps when computing the alternative representation for the given time step.
- the convolutional subnetwork can store the output values of the computation the first time that the computation is performed and then re-use the stored output values at subsequent time steps.
- the first convolutional layer in the convolutional subnetwork will apply the same filter or filters multiple times to the same audio sample values during the generation of an audio sequence. Rather than re-compute the output of these filter applications at each time step, the system can re-use outputs computed at previous time steps.
- one stack of dilated causal convolutional layers with a very large (long) receptive field, but preferably fewer units per layer, may be employed to condition another (larger) stack with a smaller receptive field.
- the larger stack may process a shorter part of the audio signal, for example cropped at the end.
- Embodiments of the subject matter and the functional operations described in this specification can be implemented in digital electronic circuitry, in tangibly-embodied computer software or firmware, in computer hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them.
- Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non transitory storage medium for execution by, or to control the operation of, data processing apparatus.
- the computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.
- the program instructions can be encoded on an artificially generated propagated signal, e.g., a machine-generated electrical, optical, or
- electromagnetic signal that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus.
- data processing apparatus refers to data processing hardware and encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable processor, a computer, or multiple processors or computers.
- the apparatus can also be, or further include, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit).
- the apparatus can optionally include, in addition to hardware, code that creates an execution environment for computer programs, e.g., code that constitutes processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them.
- a computer program which may also be referred to or described as a program, software, a software application, an app, a module, a software module, a script, or code, can be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages; and it can be deployed in any form, including as a stand alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.
- a program may, but need not, correspond to a file in a file system.
- a program can be stored in a portion of a file that holds other programs or data, e.g., one or more scripts stored in a markup language document, in a single file dedicated to the program in question, or in multiple coordinated files, e.g., files that store one or more modules, sub programs, or portions of code.
- a computer program can be deployed to be executed on one computer or on multiple computers that are located at one site or distributed across multiple sites and interconnected by a data communication network.
- the term “database” is used broadly to refer to any collection of data: the data does not need to be structured in any particular way, or structured at all, and it can be stored on storage devices in one or more locations.
- the index database can include multiple collections of data, each of which may be organized and accessed differently.
- engine is used broadly to refer to a software-based system, subsystem, or process that is programmed to perform one or more specific functions.
- an engine will be implemented as one or more software modules or components, installed on one or more computers in one or more locations. In some cases, one or more computers will be dedicated to a particular engine; in other cases, multiple engines can be installed and running on the same computer or computers.
- the processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data and generating output.
- the processes and logic flows can also be performed by special purpose logic circuitry, e.g., an FPGA or an ASIC, or by a combination of special purpose logic circuitry and one or more programmed computers.
- Computers suitable for the execution of a computer program can be based on general or special purpose microprocessors or both, or any other kind of central processing unit.
- a central processing unit will receive instructions and data from a read only memory or a random access memory or both.
- the essential elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices for storing instructions and data.
- the central processing unit and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.
- a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto optical disks, or optical disks. However, a computer need not have such devices.
- a computer can be embedded in another device, e.g., a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device, e.g., a universal serial bus (USB) flash drive, to name just a few.
- PDA personal digital assistant
- GPS Global Positioning System
- USB universal serial bus
- Computer readable media suitable for storing computer program instructions and data include all forms of non volatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto optical disks; and CD ROM and DVD-ROM disks.
- semiconductor memory devices e.g., EPROM, EEPROM, and flash memory devices
- magnetic disks e.g., internal hard disks or removable disks
- magneto optical disks e.g., CD ROM and DVD-ROM disks.
- embodiments of the subject matter described in this specification can be implemented on a computer having a display device, e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user and a keyboard and a pointing device, e.g., a mouse or a trackball, by which the user can provide input to the computer.
- a display device e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor
- keyboard and a pointing device e.g., a mouse or a trackball
- Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input.
- a computer can interact with a user by sending documents to and receiving documents from a device that is used by the user; for example, by sending web pages to a web browser on a user's device in response to requests received from the web browser.
- a computer can interact with a user by sending text messages or other forms of message to a personal device, e.g., a smartphone that is running a messaging application, and receiving responsive messages from the user in return.
- Data processing apparatus for implementing machine learning models can also include, for example, special-purpose hardware accelerator units for processing common and compute-intensive parts of machine learning training or production, i.e., inference, workloads.
- Machine learning models can be implemented and deployed using a machine learning framework, .e.g., a TensorFlow framework, a Microsoft Cognitive Toolkit framework, an Apache Singa framework, or an Apache MXNet framework.
- a machine learning framework .e.g., a TensorFlow framework, a Microsoft Cognitive Toolkit framework, an Apache Singa framework, or an Apache MXNet framework.
- Embodiments of the subject matter described in this specification can be implemented in a computing system that includes a back end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front end component, e.g., a client computer having a graphical user interface, a web browser, or an app through which a user can interact with an implementation of the subject matter described in this specification, or any combination of one or more such back end, middleware, or front end components.
- the components of the system can be interconnected by any form or medium of digital data communication,
- Examples of communication networks include a local area network (LAN) and a wide area network (WAN), e.g., the Internet.
- LAN local area network
- WAN wide area network
- the computing system can include clients and servers.
- a client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.
- a server transmits data, e.g., an HTML page, to a user device, e.g., for purposes of displaying data to and receiving user input from a user interacting with the device, which acts as a client.
- Data generated at the user device e.g., a result of the user interaction, can be received at the server from the device.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Acoustics & Sound (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Electrically Operated Instructional Devices (AREA)
- Complex Calculations (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
- Stereophonic System (AREA)
Abstract
Description
Claims
Priority Applications (14)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011082855.5A CN112289342B (en) | 2016-09-06 | 2017-09-06 | Generating audio using neural networks |
JP2019522236A JP6577159B1 (en) | 2016-09-06 | 2017-09-06 | Generating audio using neural networks |
BR112019004524-4A BR112019004524B1 (en) | 2016-09-06 | 2017-09-06 | NEURAL NETWORK SYSTEM, ONE OR MORE NON-TRAINER COMPUTER READABLE STORAGE MEDIA AND METHOD FOR AUTOREGRESSIVELY GENERATING AN AUDIO DATA OUTPUT SEQUENCE |
EP17780543.9A EP3497629B1 (en) | 2016-09-06 | 2017-09-06 | Generating audio using neural networks |
KR1020197009838A KR102353284B1 (en) | 2016-09-06 | 2017-09-06 | Generate audio using neural networks |
CN201780065523.6A CN109891434B (en) | 2016-09-06 | 2017-09-06 | Generating audio using neural networks |
CA3036067A CA3036067C (en) | 2016-09-06 | 2017-09-06 | Generating audio using neural networks |
EP20195353.6A EP3822863B1 (en) | 2016-09-06 | 2017-09-06 | Generating audio using neural networks |
AU2017324937A AU2017324937B2 (en) | 2016-09-06 | 2017-09-06 | Generating audio using neural networks |
US16/030,742 US10304477B2 (en) | 2016-09-06 | 2018-07-09 | Generating audio using neural networks |
US16/390,549 US10803884B2 (en) | 2016-09-06 | 2019-04-22 | Generating audio using neural networks |
US17/020,348 US11386914B2 (en) | 2016-09-06 | 2020-09-14 | Generating audio using neural networks |
US17/838,985 US11869530B2 (en) | 2016-09-06 | 2022-06-13 | Generating audio using neural networks |
US18/519,986 US20240135955A1 (en) | 2016-09-06 | 2023-11-27 | Generating audio using neural networks |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US201662384115P | 2016-09-06 | 2016-09-06 | |
US62/384,115 | 2016-09-06 |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US16/030,742 Continuation US10304477B2 (en) | 2016-09-06 | 2018-07-09 | Generating audio using neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2018048934A1 true WO2018048934A1 (en) | 2018-03-15 |
Family
ID=60022154
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2017/050320 WO2018048934A1 (en) | 2016-09-06 | 2017-09-06 | Generating audio using neural networks |
Country Status (9)
Country | Link |
---|---|
US (5) | US10304477B2 (en) |
EP (2) | EP3497629B1 (en) |
JP (3) | JP6577159B1 (en) |
KR (1) | KR102353284B1 (en) |
CN (2) | CN112289342B (en) |
AU (1) | AU2017324937B2 (en) |
BR (1) | BR112019004524B1 (en) |
CA (2) | CA3155320A1 (en) |
WO (1) | WO2018048934A1 (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109771944A (en) * | 2018-12-19 | 2019-05-21 | 武汉西山艺创文化有限公司 | A kind of sound effect of game generation method, device, equipment and storage medium |
EP3719711A2 (en) | 2020-07-30 | 2020-10-07 | Institutul Roman De Stiinta Si Tehnologie | Method of detecting anomalous data, machine computing unit, computer program |
US10971170B2 (en) * | 2018-08-08 | 2021-04-06 | Google Llc | Synthesizing speech from text using neural networks |
US11024321B2 (en) | 2018-11-30 | 2021-06-01 | Google Llc | Speech coding using auto-regressive generative neural networks |
US11587552B2 (en) | 2019-04-30 | 2023-02-21 | Sutherland Global Services Inc. | Real time key conversational metrics prediction and notability |
US11781416B2 (en) | 2019-10-16 | 2023-10-10 | Saudi Arabian Oil Company | Determination of elastic properties of a geological formation using machine learning applied to data acquired while drilling |
US11796714B2 (en) | 2020-12-10 | 2023-10-24 | Saudi Arabian Oil Company | Determination of mechanical properties of a geological formation using deep learning applied to data acquired while drilling |
Families Citing this family (78)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9965247B2 (en) | 2016-02-22 | 2018-05-08 | Sonos, Inc. | Voice controlled media playback system based on user profile |
US9811314B2 (en) | 2016-02-22 | 2017-11-07 | Sonos, Inc. | Metadata exchange involving a networked playback system and a networked microphone system |
US10095470B2 (en) | 2016-02-22 | 2018-10-09 | Sonos, Inc. | Audio response playback |
US9772817B2 (en) | 2016-02-22 | 2017-09-26 | Sonos, Inc. | Room-corrected voice detection |
US10264030B2 (en) | 2016-02-22 | 2019-04-16 | Sonos, Inc. | Networked microphone device control |
US9947316B2 (en) | 2016-02-22 | 2018-04-17 | Sonos, Inc. | Voice control of a media playback system |
US9978390B2 (en) | 2016-06-09 | 2018-05-22 | Sonos, Inc. | Dynamic player selection for audio signal processing |
US10134399B2 (en) | 2016-07-15 | 2018-11-20 | Sonos, Inc. | Contextualization of voice inputs |
US10115400B2 (en) | 2016-08-05 | 2018-10-30 | Sonos, Inc. | Multiple voice services |
EP3497629B1 (en) | 2016-09-06 | 2020-11-04 | Deepmind Technologies Limited | Generating audio using neural networks |
US11080591B2 (en) * | 2016-09-06 | 2021-08-03 | Deepmind Technologies Limited | Processing sequences using convolutional neural networks |
WO2018048945A1 (en) * | 2016-09-06 | 2018-03-15 | Deepmind Technologies Limited | Processing sequences using convolutional neural networks |
US9942678B1 (en) | 2016-09-27 | 2018-04-10 | Sonos, Inc. | Audio playback settings for voice interaction |
US10181323B2 (en) | 2016-10-19 | 2019-01-15 | Sonos, Inc. | Arbitration-based voice recognition |
CN110023963B (en) | 2016-10-26 | 2023-05-30 | 渊慧科技有限公司 | Processing text sequences using neural networks |
EP3745394B1 (en) * | 2017-03-29 | 2023-05-10 | Google LLC | End-to-end text-to-speech conversion |
US10475449B2 (en) | 2017-08-07 | 2019-11-12 | Sonos, Inc. | Wake-word detection suppression |
KR102410820B1 (en) * | 2017-08-14 | 2022-06-20 | 삼성전자주식회사 | Method and apparatus for recognizing based on neural network and for training the neural network |
JP7209275B2 (en) * | 2017-08-31 | 2023-01-20 | 国立研究開発法人情報通信研究機構 | AUDIO DATA LEARNING DEVICE, AUDIO DATA REASONING DEVICE, AND PROGRAM |
US10048930B1 (en) | 2017-09-08 | 2018-08-14 | Sonos, Inc. | Dynamic computation of system response volume |
US10446165B2 (en) | 2017-09-27 | 2019-10-15 | Sonos, Inc. | Robust short-time fourier transform acoustic echo cancellation during audio playback |
US10482868B2 (en) | 2017-09-28 | 2019-11-19 | Sonos, Inc. | Multi-channel acoustic echo cancellation |
US10621981B2 (en) | 2017-09-28 | 2020-04-14 | Sonos, Inc. | Tone interference cancellation |
US10466962B2 (en) | 2017-09-29 | 2019-11-05 | Sonos, Inc. | Media playback system with voice assistance |
US10880650B2 (en) | 2017-12-10 | 2020-12-29 | Sonos, Inc. | Network microphone devices with automatic do not disturb actuation capabilities |
JP7082357B2 (en) * | 2018-01-11 | 2022-06-08 | ネオサピエンス株式会社 | Text-to-speech synthesis methods using machine learning, devices and computer-readable storage media |
WO2019152722A1 (en) | 2018-01-31 | 2019-08-08 | Sonos, Inc. | Device designation of playback and network microphone device arrangements |
US11175880B2 (en) | 2018-05-10 | 2021-11-16 | Sonos, Inc. | Systems and methods for voice-assisted media content selection |
US10959029B2 (en) | 2018-05-25 | 2021-03-23 | Sonos, Inc. | Determining and adapting to changes in microphone performance of playback devices |
CA3103470A1 (en) | 2018-06-12 | 2019-12-19 | Intergraph Corporation | Artificial intelligence applications for computer-aided dispatch systems |
US10681460B2 (en) | 2018-06-28 | 2020-06-09 | Sonos, Inc. | Systems and methods for associating playback devices with voice assistant services |
US10461710B1 (en) | 2018-08-28 | 2019-10-29 | Sonos, Inc. | Media playback system with maximum volume setting |
US11076035B2 (en) | 2018-08-28 | 2021-07-27 | Sonos, Inc. | Do not disturb feature for audio notifications |
US10587430B1 (en) | 2018-09-14 | 2020-03-10 | Sonos, Inc. | Networked devices, systems, and methods for associating playback devices based on sound codes |
US11024331B2 (en) | 2018-09-21 | 2021-06-01 | Sonos, Inc. | Voice detection optimization using sound metadata |
US10811015B2 (en) | 2018-09-25 | 2020-10-20 | Sonos, Inc. | Voice detection optimization based on selected voice assistant service |
CN112789625A (en) | 2018-09-27 | 2021-05-11 | 渊慧科技有限公司 | Committed information rate variational self-encoder |
US11100923B2 (en) | 2018-09-28 | 2021-08-24 | Sonos, Inc. | Systems and methods for selective wake word detection using neural network models |
US11899519B2 (en) | 2018-10-23 | 2024-02-13 | Sonos, Inc. | Multiple stage network microphone device with reduced power consumption and processing load |
US10977872B2 (en) | 2018-10-31 | 2021-04-13 | Sony Interactive Entertainment Inc. | Graphical style modification for video games using machine learning |
US11636673B2 (en) | 2018-10-31 | 2023-04-25 | Sony Interactive Entertainment Inc. | Scene annotation using machine learning |
US11375293B2 (en) | 2018-10-31 | 2022-06-28 | Sony Interactive Entertainment Inc. | Textual annotation of acoustic effects |
US10854109B2 (en) | 2018-10-31 | 2020-12-01 | Sony Interactive Entertainment Inc. | Color accommodation for on-demand accessibility |
EP3654249A1 (en) * | 2018-11-15 | 2020-05-20 | Snips | Dilated convolutions and gating for efficient keyword spotting |
US11183183B2 (en) | 2018-12-07 | 2021-11-23 | Sonos, Inc. | Systems and methods of operating media playback systems having multiple voice assistant services |
US11132989B2 (en) | 2018-12-13 | 2021-09-28 | Sonos, Inc. | Networked microphone devices, systems, and methods of localized arbitration |
US10602268B1 (en) | 2018-12-20 | 2020-03-24 | Sonos, Inc. | Optimization of network microphone devices using noise classification |
JP7192882B2 (en) * | 2018-12-26 | 2022-12-20 | 日本電信電話株式会社 | Speech rhythm conversion device, model learning device, methods therefor, and program |
US11315556B2 (en) | 2019-02-08 | 2022-04-26 | Sonos, Inc. | Devices, systems, and methods for distributed voice processing by transmitting sound data associated with a wake word to an appropriate device for identification |
US10867604B2 (en) | 2019-02-08 | 2020-12-15 | Sonos, Inc. | Devices, systems, and methods for distributed voice processing |
US11120794B2 (en) | 2019-05-03 | 2021-09-14 | Sonos, Inc. | Voice assistant persistence across multiple network microphone devices |
CN110136731B (en) * | 2019-05-13 | 2021-12-24 | 天津大学 | Cavity causal convolution generation confrontation network end-to-end bone conduction voice blind enhancement method |
CN113874934A (en) * | 2019-05-23 | 2021-12-31 | 谷歌有限责任公司 | Variational embedding capacity in expressive end-to-end speech synthesis |
US11200894B2 (en) | 2019-06-12 | 2021-12-14 | Sonos, Inc. | Network microphone device with command keyword eventing |
US10586540B1 (en) | 2019-06-12 | 2020-03-10 | Sonos, Inc. | Network microphone device with command keyword conditioning |
US11361756B2 (en) | 2019-06-12 | 2022-06-14 | Sonos, Inc. | Conditional wake word eventing based on environment |
US10871943B1 (en) | 2019-07-31 | 2020-12-22 | Sonos, Inc. | Noise classification for event detection |
US11138969B2 (en) | 2019-07-31 | 2021-10-05 | Sonos, Inc. | Locally distributed keyword detection |
US11138975B2 (en) | 2019-07-31 | 2021-10-05 | Sonos, Inc. | Locally distributed keyword detection |
CN110728991B (en) * | 2019-09-06 | 2022-03-01 | 南京工程学院 | Improved recording equipment identification algorithm |
US11189286B2 (en) | 2019-10-22 | 2021-11-30 | Sonos, Inc. | VAS toggle based on device orientation |
KR20210048310A (en) | 2019-10-23 | 2021-05-03 | 삼성전자주식회사 | Electronic apparatus and control method thereof |
KR102556096B1 (en) * | 2019-11-29 | 2023-07-18 | 한국전자통신연구원 | Apparatus and method for encoding/decoding audio signal using information of previous frame |
US11200900B2 (en) | 2019-12-20 | 2021-12-14 | Sonos, Inc. | Offline voice control |
US11373095B2 (en) * | 2019-12-23 | 2022-06-28 | Jens C. Jenkins | Machine learning multiple features of depicted item |
US11562740B2 (en) | 2020-01-07 | 2023-01-24 | Sonos, Inc. | Voice verification for media playback |
US11308958B2 (en) | 2020-02-07 | 2022-04-19 | Sonos, Inc. | Localized wakeword verification |
US20210312258A1 (en) * | 2020-04-01 | 2021-10-07 | Sony Corporation | Computing temporal convolution networks in real time |
US20210350788A1 (en) * | 2020-05-06 | 2021-11-11 | Samsung Electronics Co., Ltd. | Electronic device for generating speech signal corresponding to at least one text and operating method of the electronic device |
US11308962B2 (en) | 2020-05-20 | 2022-04-19 | Sonos, Inc. | Input detection windowing |
US11482224B2 (en) | 2020-05-20 | 2022-10-25 | Sonos, Inc. | Command keywords with input detection windowing |
US11727919B2 (en) | 2020-05-20 | 2023-08-15 | Sonos, Inc. | Memory allocation for keyword spotting engines |
US11698771B2 (en) | 2020-08-25 | 2023-07-11 | Sonos, Inc. | Vocal guidance engines for playback devices |
US11984123B2 (en) | 2020-11-12 | 2024-05-14 | Sonos, Inc. | Network device interaction by range |
CN113724683B (en) * | 2021-07-23 | 2024-03-22 | 阿里巴巴达摩院(杭州)科技有限公司 | Audio generation method, computer device and computer readable storage medium |
WO2023177145A1 (en) * | 2022-03-16 | 2023-09-21 | 삼성전자주식회사 | Electronic device and method for controlling electronic device |
WO2023219292A1 (en) * | 2022-05-09 | 2023-11-16 | 삼성전자 주식회사 | Method and device for processing audio in order to classify scene |
EP4293662A1 (en) * | 2022-06-17 | 2023-12-20 | Samsung Electronics Co., Ltd. | Method and system for personalising machine learning models |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CA2810457A1 (en) * | 2013-03-25 | 2014-09-25 | Gerald Bradley PENN | System and method for applying a convolutional neural network to speech recognition |
Family Cites Families (71)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US2810457A (en) * | 1953-04-10 | 1957-10-22 | Gen Motors Corp | Lubricator |
JPH0450121Y2 (en) | 1986-04-30 | 1992-11-26 | ||
JP2522400B2 (en) * | 1989-08-10 | 1996-08-07 | ヤマハ株式会社 | Musical sound waveform generation method |
US5377302A (en) | 1992-09-01 | 1994-12-27 | Monowave Corporation L.P. | System for recognizing speech |
AU675389B2 (en) * | 1994-04-28 | 1997-01-30 | Motorola, Inc. | A method and apparatus for converting text into audible signals using a neural network |
JP3270668B2 (en) * | 1995-10-31 | 2002-04-02 | ナショナル サイエンス カウンシル | Prosody synthesizer based on artificial neural network from text to speech |
US6357176B2 (en) * | 1997-03-19 | 2002-03-19 | Mississippi State University | Soilless sod |
JPH10333699A (en) * | 1997-06-05 | 1998-12-18 | Fujitsu Ltd | Voice recognition and voice synthesizer |
US5913194A (en) * | 1997-07-14 | 1999-06-15 | Motorola, Inc. | Method, device and system for using statistical information to reduce computation and memory requirements of a neural network based speech synthesis system |
JPH11282484A (en) * | 1998-03-27 | 1999-10-15 | Victor Co Of Japan Ltd | Voice synthesizer |
US8645137B2 (en) | 2000-03-16 | 2014-02-04 | Apple Inc. | Fast, language-independent method for user authentication by voice |
DE10018134A1 (en) * | 2000-04-12 | 2001-10-18 | Siemens Ag | Determining prosodic markings for text-to-speech systems - using neural network to determine prosodic markings based on linguistic categories such as number, verb, verb particle, pronoun, preposition etc. |
JP2002123280A (en) * | 2000-10-16 | 2002-04-26 | Seiko Epson Corp | Speech synthesis method, speech synthesizer, and recording medium recording speech synthesis processing program |
US7062437B2 (en) * | 2001-02-13 | 2006-06-13 | International Business Machines Corporation | Audio renderings for expressing non-audio nuances |
US20060064177A1 (en) | 2004-09-17 | 2006-03-23 | Nokia Corporation | System and method for measuring confusion among words in an adaptive speech recognition system |
US7747070B2 (en) * | 2005-08-31 | 2010-06-29 | Microsoft Corporation | Training convolutional neural networks on graphics processing units |
KR100832556B1 (en) * | 2006-09-22 | 2008-05-26 | (주)한국파워보이스 | Speech Recognition Methods for the Robust Distant-talking Speech Recognition System |
US8504361B2 (en) * | 2008-02-07 | 2013-08-06 | Nec Laboratories America, Inc. | Deep neural networks and methods for using same |
CA2724753A1 (en) * | 2008-05-30 | 2009-12-03 | Nokia Corporation | Method, apparatus and computer program product for providing improved speech synthesis |
FR2950713A1 (en) | 2009-09-29 | 2011-04-01 | Movea Sa | SYSTEM AND METHOD FOR RECOGNIZING GESTURES |
TWI413104B (en) * | 2010-12-22 | 2013-10-21 | Ind Tech Res Inst | Controllable prosody re-estimation system and method and computer program product thereof |
CN102651217A (en) * | 2011-02-25 | 2012-08-29 | 株式会社东芝 | Method and equipment for voice synthesis and method for training acoustic model used in voice synthesis |
EP2565667A1 (en) | 2011-08-31 | 2013-03-06 | Friedrich-Alexander-Universität Erlangen-Nürnberg | Direction of arrival estimation using watermarked audio signals and microphone arrays |
US8527276B1 (en) * | 2012-10-25 | 2013-09-03 | Google Inc. | Speech synthesis using deep neural networks |
US9230550B2 (en) * | 2013-01-10 | 2016-01-05 | Sensory, Incorporated | Speaker verification and identification using artificial neural network-based sub-phonetic unit discrimination |
US9147154B2 (en) | 2013-03-13 | 2015-09-29 | Google Inc. | Classifying resources using a deep network |
US9141906B2 (en) * | 2013-03-13 | 2015-09-22 | Google Inc. | Scoring concept terms using a deep network |
US9190053B2 (en) | 2013-03-25 | 2015-11-17 | The Governing Council Of The Univeristy Of Toronto | System and method for applying a convolutional neural network to speech recognition |
US20150032449A1 (en) * | 2013-07-26 | 2015-01-29 | Nuance Communications, Inc. | Method and Apparatus for Using Convolutional Neural Networks in Speech Recognition |
CN104681034A (en) * | 2013-11-27 | 2015-06-03 | 杜比实验室特许公司 | Audio signal processing method |
US9953634B1 (en) | 2013-12-17 | 2018-04-24 | Knowles Electronics, Llc | Passive training for automatic speech recognition |
US10275704B2 (en) | 2014-06-06 | 2019-04-30 | Google Llc | Generating representations of input sequences using neural networks |
US10181098B2 (en) | 2014-06-06 | 2019-01-15 | Google Llc | Generating representations of input sequences using neural networks |
KR102332729B1 (en) | 2014-07-28 | 2021-11-30 | 삼성전자주식회사 | Speech recognition method and apparatus, speech recognition engine generation method and apparatus based on pronounce similarity |
US9821340B2 (en) * | 2014-07-28 | 2017-11-21 | Kolo Medical Ltd. | High displacement ultrasonic transducer |
US20160035344A1 (en) * | 2014-08-04 | 2016-02-04 | Google Inc. | Identifying the language of a spoken utterance |
CN110110843B (en) | 2014-08-29 | 2020-09-25 | 谷歌有限责任公司 | Method and system for processing images |
JP6814146B2 (en) * | 2014-09-25 | 2021-01-13 | サンハウス・テクノロジーズ・インコーポレーテッド | Systems and methods for capturing and interpreting audio |
US10783900B2 (en) * | 2014-10-03 | 2020-09-22 | Google Llc | Convolutional, long short-term memory, fully connected deep neural networks |
US9824684B2 (en) | 2014-11-13 | 2017-11-21 | Microsoft Technology Licensing, Llc | Prediction-based sequence recognition |
US9542927B2 (en) * | 2014-11-13 | 2017-01-10 | Google Inc. | Method and system for building text-to-speech voice from diverse recordings |
US9607217B2 (en) * | 2014-12-22 | 2017-03-28 | Yahoo! Inc. | Generating preference indices for image content |
US11080587B2 (en) * | 2015-02-06 | 2021-08-03 | Deepmind Technologies Limited | Recurrent neural networks for data item generation |
US10403269B2 (en) * | 2015-03-27 | 2019-09-03 | Google Llc | Processing audio waveforms |
US20160343366A1 (en) * | 2015-05-19 | 2016-11-24 | Google Inc. | Speech synthesis model selection |
US9595002B2 (en) | 2015-05-29 | 2017-03-14 | Sas Institute Inc. | Normalizing electronic communications using a vector having a repeating substring as input for a neural network |
CN105096939B (en) * | 2015-07-08 | 2017-07-25 | 百度在线网络技术(北京)有限公司 | voice awakening method and device |
US9786270B2 (en) | 2015-07-09 | 2017-10-10 | Google Inc. | Generating acoustic models |
CN106375231B (en) * | 2015-07-22 | 2019-11-05 | 华为技术有限公司 | A kind of flow switching method, equipment and system |
KR102413692B1 (en) | 2015-07-24 | 2022-06-27 | 삼성전자주식회사 | Apparatus and method for caculating acoustic score for speech recognition, speech recognition apparatus and method, and electronic device |
CN105068998B (en) | 2015-07-29 | 2017-12-15 | 百度在线网络技术(北京)有限公司 | Interpretation method and device based on neural network model |
CN105321525B (en) * | 2015-09-30 | 2019-02-22 | 北京邮电大学 | A kind of system and method reducing VOIP communication resource expense |
US10733979B2 (en) | 2015-10-09 | 2020-08-04 | Google Llc | Latency constraints for acoustic modeling |
US10395118B2 (en) | 2015-10-29 | 2019-08-27 | Baidu Usa Llc | Systems and methods for video paragraph captioning using hierarchical recurrent neural networks |
WO2017083695A1 (en) * | 2015-11-12 | 2017-05-18 | Google Inc. | Generating target sequences from input sequences using partial conditioning |
US10319374B2 (en) | 2015-11-25 | 2019-06-11 | Baidu USA, LLC | Deployed end-to-end speech recognition |
CN105513591B (en) * | 2015-12-21 | 2019-09-03 | 百度在线网络技术(北京)有限公司 | The method and apparatus for carrying out speech recognition with LSTM Recognition with Recurrent Neural Network model |
US10402700B2 (en) | 2016-01-25 | 2019-09-03 | Deepmind Technologies Limited | Generating images using neural networks |
CN108780519B (en) * | 2016-03-11 | 2022-09-02 | 奇跃公司 | Structural learning of convolutional neural networks |
US10460747B2 (en) | 2016-05-10 | 2019-10-29 | Google Llc | Frequency based audio analysis using neural networks |
US9972314B2 (en) | 2016-06-01 | 2018-05-15 | Microsoft Technology Licensing, Llc | No loss-optimization for weighted transducer |
US11373672B2 (en) | 2016-06-14 | 2022-06-28 | The Trustees Of Columbia University In The City Of New York | Systems and methods for speech separation and neural decoding of attentional selection in multi-speaker environments |
US9984683B2 (en) | 2016-07-22 | 2018-05-29 | Google Llc | Automatic speech recognition using multi-dimensional models |
WO2018048945A1 (en) | 2016-09-06 | 2018-03-15 | Deepmind Technologies Limited | Processing sequences using convolutional neural networks |
EP3497629B1 (en) * | 2016-09-06 | 2020-11-04 | Deepmind Technologies Limited | Generating audio using neural networks |
US11080591B2 (en) | 2016-09-06 | 2021-08-03 | Deepmind Technologies Limited | Processing sequences using convolutional neural networks |
CN110023963B (en) | 2016-10-26 | 2023-05-30 | 渊慧科技有限公司 | Processing text sequences using neural networks |
US10049106B2 (en) | 2017-01-18 | 2018-08-14 | Xerox Corporation | Natural language generation through character-based recurrent neural networks with finite-state prior knowledge |
US11934935B2 (en) | 2017-05-20 | 2024-03-19 | Deepmind Technologies Limited | Feedforward generative neural networks |
US10726858B2 (en) | 2018-06-22 | 2020-07-28 | Intel Corporation | Neural network for speech denoising trained with deep feature losses |
US10971170B2 (en) | 2018-08-08 | 2021-04-06 | Google Llc | Synthesizing speech from text using neural networks |
-
2017
- 2017-09-06 EP EP17780543.9A patent/EP3497629B1/en active Active
- 2017-09-06 CA CA3155320A patent/CA3155320A1/en active Pending
- 2017-09-06 CA CA3036067A patent/CA3036067C/en active Active
- 2017-09-06 WO PCT/US2017/050320 patent/WO2018048934A1/en unknown
- 2017-09-06 BR BR112019004524-4A patent/BR112019004524B1/en active IP Right Grant
- 2017-09-06 AU AU2017324937A patent/AU2017324937B2/en active Active
- 2017-09-06 JP JP2019522236A patent/JP6577159B1/en active Active
- 2017-09-06 KR KR1020197009838A patent/KR102353284B1/en active IP Right Grant
- 2017-09-06 EP EP20195353.6A patent/EP3822863B1/en active Active
- 2017-09-06 CN CN202011082855.5A patent/CN112289342B/en active Active
- 2017-09-06 CN CN201780065523.6A patent/CN109891434B/en active Active
-
2018
- 2018-07-09 US US16/030,742 patent/US10304477B2/en active Active
-
2019
- 2019-04-22 US US16/390,549 patent/US10803884B2/en active Active
- 2019-08-20 JP JP2019150456A patent/JP6891236B2/en active Active
-
2020
- 2020-09-14 US US17/020,348 patent/US11386914B2/en active Active
-
2021
- 2021-05-25 JP JP2021087708A patent/JP7213913B2/en active Active
-
2022
- 2022-06-13 US US17/838,985 patent/US11869530B2/en active Active
-
2023
- 2023-11-27 US US18/519,986 patent/US20240135955A1/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CA2810457A1 (en) * | 2013-03-25 | 2014-09-25 | Gerald Bradley PENN | System and method for applying a convolutional neural network to speech recognition |
Non-Patent Citations (4)
Title |
---|
AÄRON VAN DEN OORD ET AL: "Pixel Recurrent Neural Networks", 19 August 2016 (2016-08-19), pages 1 - 11, XP055435831, Retrieved from the Internet <URL:https://arxiv.org/pdf/1601.06759.pdf> [retrieved on 20171218] * |
AARON VAN DEN OORD ET AL: "WaveNet: A Generative Model for Raw Audio", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 12 September 2016 (2016-09-12), XP080725972 * |
DOWNLOAD; ZEN; HEIGA; SENIOR; ANDREW; SCHUSTER; MIKE: "Statistical parametric speech synthesis using deep neural networks", PROC. ICASSP, 2013, pages 7962 - 7966, XP032508982, DOI: doi:10.1109/ICASSP.2013.6639215 |
ZEN; HEIGA, AN EXAMPLE OF CONTEXT-DEPENDENT LABEL FORMAT FOR HMM-BASED SPEECH SYNTHESIS IN ENGLISH, 2006, Retrieved from the Internet <URL:http://hts.sp.nitech.ac.jp/?> |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10971170B2 (en) * | 2018-08-08 | 2021-04-06 | Google Llc | Synthesizing speech from text using neural networks |
US11024321B2 (en) | 2018-11-30 | 2021-06-01 | Google Llc | Speech coding using auto-regressive generative neural networks |
US11676613B2 (en) | 2018-11-30 | 2023-06-13 | Google Llc | Speech coding using auto-regressive generative neural networks |
CN109771944A (en) * | 2018-12-19 | 2019-05-21 | 武汉西山艺创文化有限公司 | A kind of sound effect of game generation method, device, equipment and storage medium |
CN109771944B (en) * | 2018-12-19 | 2022-07-12 | 武汉西山艺创文化有限公司 | Game sound effect generation method, device, equipment and storage medium |
US11587552B2 (en) | 2019-04-30 | 2023-02-21 | Sutherland Global Services Inc. | Real time key conversational metrics prediction and notability |
US11781416B2 (en) | 2019-10-16 | 2023-10-10 | Saudi Arabian Oil Company | Determination of elastic properties of a geological formation using machine learning applied to data acquired while drilling |
EP3719711A2 (en) | 2020-07-30 | 2020-10-07 | Institutul Roman De Stiinta Si Tehnologie | Method of detecting anomalous data, machine computing unit, computer program |
US11796714B2 (en) | 2020-12-10 | 2023-10-24 | Saudi Arabian Oil Company | Determination of mechanical properties of a geological formation using deep learning applied to data acquired while drilling |
Also Published As
Publication number | Publication date |
---|---|
CN109891434B (en) | 2020-10-30 |
EP3497629A1 (en) | 2019-06-19 |
CN109891434A (en) | 2019-06-14 |
US11386914B2 (en) | 2022-07-12 |
JP2021152664A (en) | 2021-09-30 |
KR20190042730A (en) | 2019-04-24 |
US20220319533A1 (en) | 2022-10-06 |
US20240135955A1 (en) | 2024-04-25 |
BR112019004524A2 (en) | 2019-05-28 |
JP6891236B2 (en) | 2021-06-18 |
JP7213913B2 (en) | 2023-01-27 |
EP3822863A1 (en) | 2021-05-19 |
AU2017324937A1 (en) | 2019-03-28 |
AU2017324937B2 (en) | 2019-12-19 |
BR112019004524B1 (en) | 2023-11-07 |
EP3822863B1 (en) | 2022-11-02 |
CA3036067C (en) | 2023-08-01 |
KR102353284B1 (en) | 2022-01-19 |
CN112289342B (en) | 2024-03-19 |
US20180322891A1 (en) | 2018-11-08 |
JP2020003809A (en) | 2020-01-09 |
CA3036067A1 (en) | 2018-03-15 |
US20190251987A1 (en) | 2019-08-15 |
US10304477B2 (en) | 2019-05-28 |
JP6577159B1 (en) | 2019-09-18 |
US11869530B2 (en) | 2024-01-09 |
US20200411032A1 (en) | 2020-12-31 |
US10803884B2 (en) | 2020-10-13 |
CA3155320A1 (en) | 2018-03-15 |
EP3497629B1 (en) | 2020-11-04 |
CN112289342A (en) | 2021-01-29 |
JP2019532349A (en) | 2019-11-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11869530B2 (en) | Generating audio using neural networks | |
US11948066B2 (en) | Processing sequences using convolutional neural networks | |
US11069345B2 (en) | Speech recognition using convolutional neural networks | |
US11481646B2 (en) | Selecting answer spans from electronic documents using neural networks | |
US20200410344A1 (en) | Fast decoding in sequence models using discrete latent variables | |
US20210089909A1 (en) | High fidelity speech synthesis with adversarial networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 17780543 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 3036067 Country of ref document: CA Ref document number: 2019522236 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 2017780543 Country of ref document: EP Effective date: 20190313 |
|
REG | Reference to national code |
Ref country code: BR Ref legal event code: B01A Ref document number: 112019004524 Country of ref document: BR |
|
ENP | Entry into the national phase |
Ref document number: 2017324937 Country of ref document: AU Date of ref document: 20170906 Kind code of ref document: A |
|
ENP | Entry into the national phase |
Ref document number: 20197009838 Country of ref document: KR Kind code of ref document: A |
|
ENP | Entry into the national phase |
Ref document number: 112019004524 Country of ref document: BR Kind code of ref document: A2 Effective date: 20190307 |