Generative adversarial network-based speech bandwidth extender and extension method
AssigneeIucf-Hyu (Industry-University Cooperation Foundation Hanyang University)
InventorsJoon-Hyuk Chang, Kyoungjin Noh
Life of this patent
Estimating the term…
Abstract
Proposed are a generative adversarial network-based speech bandwidth extender and extension method. A generative adversarial network-based speech bandwidth extension method, according to an embodiment, comprises the steps of: extracting feature vectors from a narrowband (NB) signal and a wideband (WB) signal of a speech; estimating the feature vector of the wideband signal from the feature vector of the narrowband signal; and learning a deep neural network classification model for discriminating the estimated feature vector of the wideband signal from the actually extracted feature vector of the wideband signal and the actually extracted feature vector of the narrowband signal.
Claims12
A method for generative adversarial network-based speech bandwidth extension, the method comprising:
- extracting feature vectors from a narrowband (NB) signal and a wideband (WB) signal of a speech;
- training a deep neural network (DNN) generation model that estimates a feature vector of the wideband signal using the extracted feature vector of the narrowband signal as an input;
- training a DNN classification model that discriminates the estimated feature vector of the wideband signal from the extracted feature vector of the wideband signal;
- and
- retraining the DNN generation model such that the estimated feature vector of the wideband signal is misidentified as the extracted feature vector of the wideband signal by the trained DNN classification model.
An apparatus for generative adversarial network-based speech bandwidth extension, the apparatus comprising:
- a feature vector extractor configured to extract feature vectors from a narrowband (NB) signal and a wideband (WB) signal of a speech;
- a generation modeler configured to train a deep neural network (DNN) generation model that estimates a feature vector of the wideband signal using the extracted feature vector of the narrowband signal as an input;
- and
- a classification modeler configured to train a DNN classification model that discriminates the estimated feature vector of the wideband signal from the extracted feature vector of the wideband signal,
- wherein the generation modeler is configured to retrain the DNN generation model such that the estimated feature vector of the wideband signal is misidentified as the extracted feature vector of the wideband signal by the trained DNN classification model.
Dependencies are inferred from the claim text.