Improving Unsupervised Subword Modeling Via Disentangled Speech Representation Learning And Transformation
2019 · Siyuan Feng, Tan Lee
Abstract
This study tackles unsupervised subword modeling in the zero-resource scenario, learning frame-level speech representation that is phonetically discriminative and speaker-invariant, using only untranscribed speech for target languages. Frame label acquisition is an essential step in solving this problem. High quality frame labels should be in good consistency with golden transcriptions and robust to speaker variation. We propose to improve frame label acquisition in our previously adopted deep neural network-bottleneck feature (DNN-BNF) architecture by applying the factorized hierarchical variational autoencoder (FHVAE). FHVAEs learn to disentangle linguistic content and speaker identity information encoded in speech. By discarding or unifying speaker information, speaker-invariant features are learned and fed as inputs to DPGMM frame clustering and DNN-BNF training. Experiments conducted on ZeroSpeech 2017 show that our proposed approaches achieve \(2.4%\) and \(0.6%\) absolute ABX er
Authors
(none)
Tags
Stats
Related papers
- Combining Adversarial Training And Disentangled Speech Representation For Robust Zero-resource Subword Modeling (2019)7.16
- Exploiting Cross-lingual Speaker And Phonetic Diversity For Unsupervised Subword Modeling (2019)6.77
- Multilingual And Unsupervised Subword Modeling For Zero-resource Languages (2018)7.81
- Unsupervised Neural And Bayesian Models For Zero-resource Speech Processing (2017)0.00
- Robust Disentangled Variational Speech Representation Learning For Zero-shot Voice Conversion (2022)10.97
- The Effectiveness Of Unsupervised Subword Modeling With Autoregressive And Cross-lingual Phone-aware Networks (2020)2.26
- Unsupervised Representation Learning Of Speech For Dialect Identification (2018)7.16
- Unsupervised Acoustic Unit Discovery For Speech Synthesis Using Discrete Latent-variable Neural Networks (2019)9.59