Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Who is Authentic Speaker

Qiang Huang

TL;DR

The paper tackles authentic speaker recognition from voices transformed by deep learning VC, addressing risks of deception and privacy concerns. It couples an encoder–decoder voice-conversion pipeline (FragmentVC) with a speaker recognizer that uses hierarchical VLAD to produce fixed-length descriptors from variable-length spectrogram features. The study shows that hierarchical VLAD with VLAD clustering outperforms single-VLAD baselines, with 64 clusters yielding the best Top-1 performance under constrained target utterances, and provides insights on how target utterances influence converted voice quality and recognition difficulty. This work presents a feasible framework for validating speaker authenticity under voice conversion and points to future improvements via attention mechanisms, speech separation, and benchmark datasets.

Abstract

Voice conversion (VC) using deep learning technologies can now generate high quality one-to-many voices and thus has been used in some practical application fields, such as entertainment and healthcare. However, voice conversion can pose potential social issues when manipulated voices are employed for deceptive purposes. Moreover, it is a big challenge to find who are real speakers from the converted voices as the acoustic characteristics of source speakers are changed greatly. In this paper we attempt to explore the feasibility of identifying authentic speakers from converted voices. This study is conducted with the assumption that certain information from the source speakers persists, even when their voices undergo conversion into different target voices. Therefore our experiments are geared towards recognising the source speakers given the converted voices, which are generated by using FragmentVC on the randomly paired utterances from source and target speakers. To improve the robustness against converted voices, our recognition model is constructed by using hierarchical vector of locally aggregated descriptors (VLAD) in deep neural networks. The authentic speaker recognition system is mainly tested in two aspects, including the impact of quality of converted voices and the variations of VLAD. The dataset used in this work is VCTK corpus, where source and target speakers are randomly paired. The results obtained on the converted utterances show promising performances in recognising authentic speakers from converted voices.

Who is Authentic Speaker

TL;DR

The paper tackles authentic speaker recognition from voices transformed by deep learning VC, addressing risks of deception and privacy concerns. It couples an encoder–decoder voice-conversion pipeline (FragmentVC) with a speaker recognizer that uses hierarchical VLAD to produce fixed-length descriptors from variable-length spectrogram features. The study shows that hierarchical VLAD with VLAD clustering outperforms single-VLAD baselines, with 64 clusters yielding the best Top-1 performance under constrained target utterances, and provides insights on how target utterances influence converted voice quality and recognition difficulty. This work presents a feasible framework for validating speaker authenticity under voice conversion and points to future improvements via attention mechanisms, speech separation, and benchmark datasets.

Abstract

Voice conversion (VC) using deep learning technologies can now generate high quality one-to-many voices and thus has been used in some practical application fields, such as entertainment and healthcare. However, voice conversion can pose potential social issues when manipulated voices are employed for deceptive purposes. Moreover, it is a big challenge to find who are real speakers from the converted voices as the acoustic characteristics of source speakers are changed greatly. In this paper we attempt to explore the feasibility of identifying authentic speakers from converted voices. This study is conducted with the assumption that certain information from the source speakers persists, even when their voices undergo conversion into different target voices. Therefore our experiments are geared towards recognising the source speakers given the converted voices, which are generated by using FragmentVC on the randomly paired utterances from source and target speakers. To improve the robustness against converted voices, our recognition model is constructed by using hierarchical vector of locally aggregated descriptors (VLAD) in deep neural networks. The authentic speaker recognition system is mainly tested in two aspects, including the impact of quality of converted voices and the variations of VLAD. The dataset used in this work is VCTK corpus, where source and target speakers are randomly paired. The results obtained on the converted utterances show promising performances in recognising authentic speakers from converted voices.
Paper Structure (9 sections, 1 equation, 5 figures, 3 tables)

This paper contains 9 sections, 1 equation, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Theoretical Framework
  3. Voice Conversion
  4. Authentic Speaker Recognition using VLAD
  5. Hierarchical VLAD
  6. Data
  7. Experimental Setup
  8. Result and Analysis
  9. Conclusion and Future Work

Figures (5)

  • Figure 1: Architecture of authentic speaker recognition system, mainly including voice conversion and speaker recognition
  • Figure 2: Hierarchical VLAD for authentic speaker recognition: (a) Architecture of thin ResNet34 followed by VLAD and classifier; (b)proposed structure by linking the output of each sub-convolution block in the last convolutional block with an individual VLAD layer.
  • Figure 3: Spectrogram of source utterance and converted utterance.
  • Figure 4: Authentic speaker recognition accuracy when varying the number of VLAD clusters.
  • Figure 5: Authentic speaker recognition accuracy when using different number of target utterances