A method of encoding an audio signal and an encoder and a method of decoding an audio signal and a decoder are provided. The method of encoding an audio signal includes outputting a decoded signal by using a bitstream that encodes an audio signal, separating the decoded signal into a low-band signal and a high-band signal by using a sound source separator, upsampling the low-band signal, upsampling the high-band signal, and restoring the audio signal by synthesizing the upsampled low-band signal with the upsampled high-band signal, wherein the bitstream is generated by encoding a superimposed signal in which a signal in a high frequency band of the audio signal is superimposed on a low frequency band of the audio signal.

An audio output apparatus is disclosed. The audio output apparatus that outputs a multi-channel audio signal through a plurality of speakers disposed at different locations, the audio output apparatus includes an input interface, and a processor configured to, based on the multi-channel audio signal input through the inputter being received, obtain scene information on a type of audio included in the multi-channel audio signal and sound image angle information about an angle formed by sound image of the type of audio included in the multi-channel audio signal based on a virtual user, and generate an output signal to be output through the plurality of speakers from the multi-channel audio signal based on the obtained scene information and sound image angle information, wherein the type of audio includes at least one of sound effect, shouting sound, music, and voice, and a number of the plurality of speakers is equal to or greater than a number of channels of the multi-channel audio signal.

An audio output apparatus is disclosed. The audio output apparatus that outputs a multi-channel audio signal through a plurality of speakers disposed at different locations, the audio output apparatus includes an input interface, and a processor configured to, based on the multi-channel audio signal input through the inputter being received, obtain scene information on a type of audio included in the multi-channel audio signal and sound image angle information about an angle formed by sound image of the type of audio included in the multi-channel audio signal based on a virtual user, and generate an output signal to be output through the plurality of speakers from the multi-channel audio signal based on the obtained scene information and sound image angle information, wherein the type of audio includes at least one of sound effect, shouting sound, music, and voice, and a number of the plurality of speakers is equal to or greater than a number of channels of the multi-channel audio signal.

Embodiments of a production-quality text-to-speech (TTS) system constructed from deep neural networks are described. System embodiments comprise five major building blocks: a segmentation model for locating phoneme boundaries, a grapheme-to-phoneme conversion model, a phoneme duration prediction model, a fundamental frequency prediction model, and an audio synthesis model. For embodiments of the segmentation model, phoneme boundary detection was performed with deep neural networks using Connectionist Temporal Classification (CTC) loss. For embodiments of the audio synthesis model, a variant of WaveNet was created that requires fewer parameters and trains faster than the original. By using a neural network for each component, system embodiments are simpler and more flexible than traditional TTS systems, where each component requires laborious feature engineering and extensive domain expertise. Inference with system embodiments may be performed faster than real time.

Embodiments of a production-quality text-to-speech (TTS) system constructed from deep neural networks are described. System embodiments comprise five major building blocks: a segmentation model for locating phoneme boundaries, a grapheme-to-phoneme conversion model, a phoneme duration prediction model, a fundamental frequency prediction model, and an audio synthesis model. For embodiments of the segmentation model, phoneme boundary detection was performed with deep neural networks using Connectionist Temporal Classification (CTC) loss. For embodiments of the audio synthesis model, a variant of WaveNet was created that requires fewer parameters and trains faster than the original. By using a neural network for each component, system embodiments are simpler and more flexible than traditional TTS systems, where each component requires laborious feature engineering and extensive domain expertise. Inference with system embodiments may be performed faster than real time.

A speech synthesizer for evaluating quality of a synthesized speech using artificial intelligence includes a database configured to store a synthesized speech corresponding to text, a correct speech corresponding to the text and a speech quality evaluation model for evaluating the quality of the synthesized speech, and a processor configured to compare a first speech feature set indicating a feature of the synthesized speech and a second speech feature set indicating a feature of the correct speech, acquire a quality evaluation index set including indices used to evaluate the quality of the synthesized speech according to a result of comparison, and determine weights as model parameters of the speech quality evaluation model using the acquired quality evaluation index set and the speech quality evaluation model.

A speech synthesizer for evaluating quality of a synthesized speech using artificial intelligence includes a database configured to store a synthesized speech corresponding to text, a correct speech corresponding to the text and a speech quality evaluation model for evaluating the quality of the synthesized speech, and a processor configured to compare a first speech feature set indicating a feature of the synthesized speech and a second speech feature set indicating a feature of the correct speech, acquire a quality evaluation index set including indices used to evaluate the quality of the synthesized speech according to a result of comparison, and determine weights as model parameters of the speech quality evaluation model using the acquired quality evaluation index set and the speech quality evaluation model.

Systems, methods and computer-readable media are provided for speech enhancement using a hybrid neural network. An example process can include receiving, by a first neural network portion of the hybrid neural network, audio data and reference data, the audio data including speech data, noise data, and echo data; filtering, by the first neural network portion, a portion of the audio data based on adapted coefficients of the first neural network portion, the portion of the audio data including the noise data and/or echo data; based on the filtering, generating, by the first neural network portion, filtered audio data including the speech data and an unfiltered portion of the noise data and/or echo data; and based on the filtered audio data and the reference data, extracting, by a second neural network portion of the hybrid neural network, the speech data from the filtered audio data.

Systems, methods and computer-readable media are provided for speech enhancement using a hybrid neural network. An example process can include receiving, by a first neural network portion of the hybrid neural network, audio data and reference data, the audio data including speech data, noise data, and echo data; filtering, by the first neural network portion, a portion of the audio data based on adapted coefficients of the first neural network portion, the portion of the audio data including the noise data and/or echo data; based on the filtering, generating, by the first neural network portion, filtered audio data including the speech data and an unfiltered portion of the noise data and/or echo data; and based on the filtered audio data and the reference data, extracting, by a second neural network portion of the hybrid neural network, the speech data from the filtered audio data.

Systems and methods for generating real-time synthetic crowd responses for events, to augment the experience of event participants, remote viewers, and the like. Various sensors monitor the event in question, and various event properties are derived from their output using an event state model. These event properties, along with various event parameters such as score, time remaining, etc., are then input to a machine learning model that determines a real-time synthetic audience reaction tailored to the immediate state of the event. Reaction parameters are used to generate a corresponding crowd or audience audio signal, which may be broadcast to event participants, viewers, spectators, or anyone who may be interested. This instantaneous, realistic crowd reaction more closely simulates the experience of events with full on-site audiences, enhancing the viewing experience of both event participants and those watching.