Real-time audio processing system, real-time audio processing program, and method for training speech analysis model
AssigneeRealtek Semiconductor Corp.
InventorYen-Hsun Chu
Life of this patent
Estimating the term…
Abstract
An audio real-time processing system, an audio real-time processing program product and method for training speech analysis model are provided. The speech analysis model is firstly trained to obtain, from an original audio, mask information which is used to mask the original audio to get a target audio. The system obtains a plurality of analyzed audio according to the target audio and the original audio, obtains repeated audio section according to the plurality of the analyzed and output the repeated audio section.
Claims6
A real-time audio processing system, comprising:
- a speech analysis model, configured to obtain an analysis result and hidden layer state information according to an original audio;
- and
- a processor, configured to:
- perform a plurality of operations on the original audio by using a separator, a window, a shifting length, and the speech analysis model, obtain a plurality of target analyzed sub-audios, perform volume adjustment on each of the target analyzed sub-audios to obtain a first analyzed audio, a second analyzed audio, a third analyzed audio, a fourth analyzed audio, and the hidden layer state information, wherein the window length is 4 times the shifting length, and the first analyzed audio, the second analyzed audio, the third analyzed audio, and the fourth analyzed audio respectively include four sub-audios;
- obtain a repeated audio section according to a sub-audio that is commonly overlapped among the first analyzed audio, the second analyzed audio, the third analyzed audio, and the fourth analyzed audio;
- and
- output the repeated audio section.
A non-transitory computer-readable recording medium, comprising a real-time audio processing program, wherein the real-time audio processing program is executed by a computer to perform the following steps:
- perform a plurality of operations on an original audio by using a separator, a window, a shifting length, and a speech analysis model, obtains a plurality of target analyzed sub-audios, performs volume adjustment on each of the target analyzed sub-audios to obtain a first analyzed audio, a second analyzed audio, a third analyzed audio, a fourth analyzed audio, and hidden layer state information, wherein the window length is 4 times the shifting length, and the first analyzed audio, the second analyzed audio, the third analyzed audio, and the fourth analyzed audio respectively include four sub-audios;
- and
- obtain a repeated audio section according to a sub-audio that is commonly overlapped among the first analyzed audio, the second analyzed audio, the third analyzed audio, and the fourth analyzed audio.
Dependencies are inferred from the claim text.