Low-delay voice conversion based on maximum likelihood estimation of spectral parameter trajectory

By Takashi Muramatsu, Yamato Ohtani, Tomoki Toda, Hiroshi Saruwatari, Kiyohiro Shikano. Interspeech, 2008

1,064 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

View Paper: https://www.semanticscholar.org/paper/ce8170a83efce940b6574e3aab02b64e33fbf7c7

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Junichi Yamagishi
2. Haizhou Li
3. Xu Tan
4. Tao Qin
5. Hung-yi Lee
6. Tie-Yan Liu
7. James R. Glass
8. Wei-Ning Hsu

In Signal Processing

1. Junichi Yamagishi
2. Haizhou Li
3. Heiga Zen
4. Zhenhua Ling
5. Berrak Sisman
6. Tomoki Toda
7. Hung-yi Lee
8. James R. Glass

Citations in the corpus

• Deepfakes generation and detection: state-of-the-art, open challenges, countermeasures, and way forward (2021, methodology): “Hua et al. [291] Table 16 An overview of audio deepfake detection techniques Author Technique Features Best Evaluation performance Dataset Limitations Hand-crafted features Li et al. [277] Res2Net CQT EER=2.502 ASVspoof2019 ▪ Needs generalization improvement Yi et al. [278] GMM/LCNN CQCC EER=19.22 (”
• Intent transfer in speech-to-speech machine translation (2012, methodology): “, [20] employ an approach using Gaussian mixture models of the joint spectral vectors of source and target speakers.”
• Machine Speech Chain with One-shot Speaker Adaptation (2018, positive): “The parameters are transformed to mimic the target speakers voice with joint density Gaussian mixture model (JD-GMM) model [26].”
• NAUTILUS: A Versatile Voice Cloning System (2020, background): “The conventional VC approach is text-dependent, i.e., it expects training data to be parallel utterances of source and target speakers [29], [30].”
• Statistical Parametric Speech Synthesis Incorporating Generative Adversarial Networks (2017, methodology): “features and acoustic features more accurately than conventional hidden Markov models [5] and Gaussian mixture models [6].”
• Pretraining Techniques for Sequence-to-Sequence Voice Conversion (2020, background): “linguistic contents uttered by both source and target speakers, exists, an analysis—conversion —synthesis paradigm is often adopted [2].”
• An overview of voice conversion systems (2017, methodology): “Low-delay parameter generation algorithms without GV Muramatsu et al., 2008 ) and with GV ( Toda et al., 2012a ) have also een proposed.”
• Multi-target Voice Conversion without Parallel Data by Adversarially Learning Disentangled Audio Representations (2018, background): “In general, among the difficult problems for VC approaches, the need of aligned data [11, 12, 13, 14, 15, 16, 17], and over-smoothing of signals [1, 18, 19] are two examples carefully studied.”
• Automatic Building of Synthetic Voices from Audio Books (2010, methodology): “This voice conversion setup is based on the work done in [Toda et al., 2007], and supports the conversion considering 1) MLPG based smoothing and 2) the global variance (GV) of spectral trajectory.”
• Statistical Parametric Speech Synthesis (2007, background): “Although it works better than the post-fi ltering technique (Toda et al., 2007), it still introduces some arti fi cial sounds into synthesized speech (Zen et al., 2006b).”
• Spectral Mapping Using Artificial Neural Networks for Voice Conversion (2010, methodology): “This voice conversion setup is based on the work done in [14], and supports the conversion considering 1) the correlation between frames (referred to as MLPG) and 2) the global variance (GV) of spectral trajectory.”
• A probabilistic interpretation for artificial neural network-based voice conversion (2015, methodology): “Numerous VC approaches [1-12] have been proposed based on various models.”
• Statistical parametric speech synthesis using deep neural networks (2013, methodology): “To objectively evaluate the performance of the HMM and DNN-based systems, Mel-cepstral distortion (dB) [40], linear aperiodicity distortion (dB), voiced/unvoiced error rate (%), and root mean squared error (RMSE) in log F0 were used.”
• A Deep Generative Architecture for Postfiltering in Statistical Parametric Speech Synthesis (2015, background): “This is conceptually similar to voice conversion (VC) techniques that take into consideration the conditional probability of parallel speaker pairs [17].”
• Many-to-Many Voice Transformer Network (2020, methodology): “Examples of acoustic models include Gaussian mixture models (GMMs) [8], [9], partial least square regression [10], dynamic kernel partial least”