Optimal human-machine conversations using emotion-enhanced natural speech using hierarchical neural networks and reinforcement learning
A system and method for emotion-enhanced natural speech using dilated convolutional neural networks, wherein an audio processing server receives a raw audio waveform from a dilated convolutional artificial neural network, associates text-based emotion content markers with portions of the raw audio waveform to produce an emotion-enhanced audio waveform, and provides the emotion-enhanced audio waveform to the dilated convolutional artificial neural network for use as a new input data set.
1. A system for emotion-enhanced natural speech audio generation using dilated convolutional neural networks, comprising:
a computing device comprising a memory and a processor;
a first dilated convolutional artificial neural network stored in the memory of, and operating on the processor of, the computing device;
a second dilated convolutional artificial neural network stored in the memory of, and operating on the processor of, the computing device;
a neural network trainer, comprising a first plurality of programming instructions stored in the memory of the computing device which, when operating on the processor of the computing device, causes the computing device to:
train the first dilated convolutional artificial neural network to recognize emotion in text-based content by processing a plurality of text-based training data through the first dilated convolutional artificial neural network;
receive a first set of output data from the first dilated convolutional artificial network, the first set of output data comprising probability-based associations of text with emotions;
train the second dilated convolutional artificial neural network to recognize emotion in audio-based content by providing a plurality of audio-based training-data through the second dilated convolutional artificial neural network, the audio-based training data corresponding to the text-based training data;
receive a second set of output data from the second dilated convolutional artificial network, the second set of output data comprising probability-based associations of modulations of sounds with emotions;
construct an emotion injection model by associating text from the first set of output data with sounds from the second set of output data based on the emotions associated with each; and
an automated emotion engine comprising a second plurality of programming instructions stored in the memory of, and operating on the processor of, the computing device, wherein the programming instructions, when operating on the processor, cause the computing device to:
receive text content;
process the text content through the first dilated convolutional artificial neural network to recognize emotional states in the text content; and
convert the text content to audio content using a text-to-speech engine, modulating the audio content with the modulations of sounds associated with the text from the emotion injection model.
2. A method for emotion-enhanced natural speech audio generation using dilated convolutional neural networks, comprising:
training a first dilated convolutional artificial neural network to recognize emotion in text-based content by processing a plurality of text-based training data through the first dilated convolutional artificial neural network;
receiving a first set of output data from the first dilated convolutional artificial network, the first set of output data comprising probability-based associations of text with emotions;
training a second dilated convolutional artificial neural network to recognize emotion in audio-based content by providing a plurality of audio-based training-data through the second dilated convolutional artificial neural network, the audio-based training data corresponding to the text-based training data;
receiving a second set of output data from the second dilated convolutional artificial network, the second set of output data comprising probability-based associations of modulations of sounds with emotions;
constructing an emotion injection model by associating text from the first set of output data with sounds from the second set of output data based on the emotions associated with each;
receiving text content;
processing the text content through the first dilated convolutional artificial neural network to recognize emotional states in the text content; and
converting the text content to audio content using a text-to-speech engine, and modulating the audio content with the modulations of sounds associated with the text from the emotion injection model.