As a transcription-guided voice encoder for speaker-independent linguistic representation, we suggest Cotatron.
The multi-speaker TTS architecture that Cotatron is based on may be taught using standard TTS datasets. We develop a voice conversion system that uses Cotatron characteristics to reconstruct speech, which is comparable to earlier approaches based on Phonetic Posteriorgram (PPG).
By using 108 speakers from the VCTK dataset to train and test our system, we surpass the prior approach in terms of speaker similarity and naturalness.
Our system is also capable of converting speech from speakers who are not visible during training and using ASR to automate transcription with little performance loss.
Learn more about transcription-guided voice:
https://brainly.com/question/25703686
#SPJ4
