Speech translation and speech recognition method based on sequence dynamic compression

• Main Authors: YANG MURUN, DU QUAN
• Format: Patent
• Language: Chinese; English
• Published: 02.06.2023
• Subjects: ACOUSTICS; CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

The invention relates to a speech translation and speech recognition method based on dynamic sequence compression, and belongs to the technical field of natural language processing. The problems that in the prior art, voice data cannot be effectively compressed through a voice translation or voice recognition method, and consequently computing resources are too large are solved; or the problem of information loss caused by excessive compression of the data due to the fact that the voice data cannot be dynamically compressed step by step is solved; the speech translation method comprises the following steps: acquiring to-be-translated source language speech data; performing length prediction, dynamic compression, feature fusion and encoding on the feature sequence of the voice data through an acoustic encoder to obtain an acoustic encoder implicit vector; performing text mode conversion on the acoustic encoder implicit vector by using a text encoder, and performing feature extraction and encoding to obtain a t