BLSTM-based frequency division spectrum extension anti-noise voice conversion method

The invention discloses a BLSTM-based frequency division spectrum extension anti-noise voice conversion method. The method comprises the following specific steps of: filtering source voice and targetvoice, and extracting voice characteristic parameters including fundamental frequency, a sound channel spectrum and non-periodic components; carrying out dynamic time warping alignment on the extracted sound channel spectra of the source voice and the target voice; inputting the aligned sound channel spectra of the source voice and the target voice into a BLSTM network model of frequency divisionconversion to perform training to obtain a corresponding feature conversion network; constructing a global statistical variance consistency filtering model; filtering to-be-converted voice, extractingcharacteristic parameters of the to-be-converted voice, and preprocessing the characteristic parameters; and performing parameterized speech synthesis on the characteristic parameters of the preprocessed to-be-converted voice,