Speech synthesis in noisy environment
AssigneeLG Electronics Inc.
InventorsMinook Kim, Yongchul Park, Sungmin Han, Siyoung Yang, Sangki Kim, Juyeong Jang
Life of this patent
Estimating the term…
Abstract
Disclosed is speech synthesis in a noisy environment. According to an embodiment of the disclosure, a method of speech synthesis may generate a Lombard effect-applied synthesized speech using a feature vector generated from an utterance feature. According to the disclosure, the speech synthesis method and device may be related to artificial intelligence (AI) modules, unmanned aerial vehicles (UAVs), robots, augmented reality (AR) devices, virtual reality (VR) devices, and 5G service-related devices.
Claims18
A method of speech synthesis, the method comprising:
- obtaining speech data from a speaker;
- measuring a level of external noise;
- extracting an utterance feature from the speech data;
- generating a feature vector representing the utterance feature;
- applying the feature vector, text data, and at least one parameter for controlling speech synthesis to a pre-trained speech synthesis model including a text-to-speech synthesis model;
- and
- generating synthesized speech data using an output value from the pre-trained speech synthesis model, wherein the at least one parameter includes a parameter for adjusting a level of applying a lombard effect to the synthesized speech data using the parameter based on the level of the external noise exceeding a preset threshold.
A speech synthesis device, comprising:
- a memory;
- a transceiver communicating with an external server;
- and
- a processor configured to:
- obtain speech data from a speaker,
- measure a level of external noise,
- extract an utterance feature from the speech data,
- generate a feature vector representing the utterance feature,
- apply the feature vector, text data, and at least one parameter for controlling speech synthesis to a pre-trained speech synthesis model, and
- generate synthesized speech data using an output value from the pre-trained speech synthesis model, wherein the at least one parameter includes a parameter for adjusting a level of applying a lombard effect to the synthesized speech data using the parameter based on the level of the external noise exceeding a preset threshold.
Dependencies are inferred from the claim text.