Cross-Speaker Emotion Disentangling and Transfer for End-to-End Speech Synthesis

The cross-speaker emotion transfer task in text-to-speech (TTS) synthesis particularly aims to synthesize speech for a target speaker with the transferred from reference recorded by another (source) speaker. During process, identity information of source could also affect synthesized results, resulting issue leakage, i.e., synthetic may have voice rather than This paper proposes new method aim controllable emotional expressive and meanwhile maintain speaker’s TTS task. proposed is Tacotron2-based framework embedding as conditioning variable provide information. Two disentangling modules are contained our 1) get speaker-irrelevant emotion-discriminative embedding, 2) explicitly constrain be that expected. Moreover, we present an intuitive control strength Specifically, learned adjusted flexible scalar value, which allows controlling conveyed embedding. Extensive experiments been conducted on Mandarin disjoint corpus, results demonstrate able reasonable Compared state-of-the-art methods, gets best performance task, indicating achieves learning Furthermore, ranking test pitch trajectories plots can effectively strength, leading prosody-diverse speech.