SpeechSplit2: An efficient unsupervised speech disentanglement model for multi-aspect voice conversion

SpeechSplit is among the first algorithms that successfully disentangle speech into four components: rhythm, content, pitch, and timbre. However, the model requires exhaustive tuning of the encoder bottlenecks, which can be a daunting task and limits its generalization ability. In this work, we present SpeechSplit2, an improved version of SpeechSplit, in which simple signal processing methods are utilized to alleviate the laborious bottleneck tuning problem. We show that by feeding different inputs to each encoder, we can guide each encoder to only extract one particular aspect of speech and discard the rest, given the bottleneck size is sufficiently large to encode the corresponding information. With the same neural network architecture as SpeechSplit, SpeechSplit2 achieves comparable performance in disentangling speech components when the bottlenecks are carefully tuned and shows superior advantage over the baseline when the bottleneck size varies.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Chan, Chak Ho

Contributors dc:contributor

• Hasegawa-Johnson, Mark Allan

• Unsupervised Learning
• Voice Conversion
• Speech Disentanglement
• Signal Processing

Statement dc:rights

• Copyright 2022 Chak Ho Chan

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/117680

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F117680