Tencent Technology (Shenzhen) Company Limited

Tencent Technology Company Limited

Haolei Yuan

Xiaowei Guo

Fuzhang Wu

Binghua Qian

Weiming Dong

Xiao Mei

Xudong Zhu

Changsheng Xu

Yuqiang Ren

Xingjia Pan

This application discloses a region recognition method, apparatus, and device, and a readable storage medium, and relates to the field of artificial intelligence. The method includes: acquiring an input image, the input image including image content of a to-be-recognized region; acquiring a first rotation angle of the image content in the input image, the first rotation angle being estimated by using a recognition model; rotating a convolution kernel in the recognition model at the first rotation angle, and performing convolution processing on an image feature of the input image using the rotated convolution kernel to obtain a target feature; recognizing the target feature to obtain region data; and determining, in the input image by using the region data, a target region corresponding to the image content.

Region recognition method, apparatus and device, and readable storage medium

The present disclosure discloses a method performed at a terminal for reconstructing a speech signal, and a computer storage medium, and relates to the field of speech recognition. The method includes: collecting, by the terminal, a plurality of sound signals through a plurality of sensors of a microphone array; determining, by the terminal, a first speech signal in the plurality of sound signals; performing, by the terminal, signal separation on the first speech signal to obtain a second speech signal; and performing, by the terminal, reconstruction on the second speech signal through a distortion recovery model to obtain a reconstructed speech signal; the distortion recovery model being obtained by training based on a clean speech signal and a distorted speech signal. The embodiments of the present disclosure improve accuracy of speech recognition results.

Method and terminal for reconstructing speech signal, and computer storage medium

A speech signal processing method is performed at a terminal device, including: obtaining a recorded signal and a to-be-output speech signal, the recorded signal including a noise signal and an echo signal; calculating a loop transfer function according to the recorded signal and the speech signal; calculating a power spectrum of the echo signal and a power spectrum of the noise signal according to the recorded signal, the speech signal, and the loop transfer function; calculating a frequency weighted coefficient according to the two power spectra of the echo signal and the noise signal; adjusting a frequency amplitude of the speech signal based on the frequency weighted coefficient; and outputting the adjusted speech signal to a speaker electrically coupled to the terminal device. As such, the frequency amplitude of the speech signal is automatically adjusted according to the relative frequency distribution of a noise signal and the speech signal.

Speech signal processing method and apparatus

A method, device, and storage medium for converting text to speech are described. The method includes obtaining target text; synthesizing first machine speech corresponding to the target text; and selecting an asynchronous machine speech whose prosodic feature matches a prosodic feature of the first machine speech from an asynchronous machine speech library. The method also includes searching a synchronous machine speech library for a first synchronous machine speech corresponding to the asynchronous machine speech; synthesizing, based on a prosodic feature of the first synchronous machine speech, second machine speech corresponding to the target text; and selecting a second synchronous machine speech matching an acoustic feature of the second machine speech from the synchronous machine speech library. The method further includes splicing speaker speech units corresponding to the synchronous machine speech unit in a speaker speech library, to obtain a target speaker speech.

Speech conversion method, computer device, and storage medium

A method is performed by at least one processor, and includes acquiring training speech data by concatenating speech segments having a lowest target cost among candidate concatenation solutions, and extracting training speech segments of a first annotation type, from the training speech data, the first annotation type being used for annotating that a speech continuity of a respective one of the training speech segments is superior to a preset condition. The method further includes calculating a mean dissimilarity matrix, based on neighboring candidate speech segments corresponding to the training speech segments before concatenation, the mean dissimilarity matrix representing a mean dissimilarity in acoustic features of groups of the neighboring candidate speech segments belonging to a same type of concatenation combination relationship, and generating a concatenation cost model having a target concatenation weight, based on the mean dissimilarity matrix, the concatenation cost model corresponding to the same type of concatenation combination relationship.

Growing community of inventors

Shenzhen, China

Haolei Yuan