SONG SYNTHESIS METHOD, APPARATUS AND DEVICE, AND STORAGE MEDIUM

Disclosed is a song synthesis method, comprising: acquiring lyric recitation audio and music score information (101); by means of a preset voice recognition model and lyric pinyin text, performing duration labeling on the lyric recitation audio to obtain a recitation duration (102); analyzing initial acoustic parameters from the lyric recitation audio by means of a preset vocoder (103); according to a preset initials variable-speed dictionary, rhythm information, and beat information, extracting a singing duration from the lyric pinyin text (104); performing variable speed processing on the initial acoustic parameters according to a preset variable-speed algorithm, the recitation duration, and the singing duration (105); performing resonance peak enhancement processing on a spectrum envelope subjected to variable speed processing, so as to obtain an enhanced spectrum envelope (106); performing correction processing on the basis of pitch information, the singing duration, and a fundamental frequency subjected