Advancements in Neural Network-Based TTS (2000s): Difference between revisions

From MSc Voice Technology
Jump to navigation Jump to search
 
(One intermediate revision by the same user not shown)
Line 26: Line 26:
* These innovations have the potential to enhance the performance of speech synthesis systems, bringing them closer to natural human speech while providing greater control over the synthesized output.<ref name=":3" />
* These innovations have the potential to enhance the performance of speech synthesis systems, bringing them closer to natural human speech while providing greater control over the synthesized output.<ref name=":3" />


==== <big>DNN-Based Acoustic Modeling and Vocoder</big> ====
==== <big>DNN-Based Acoustic TTS-Modeling and Vocoder</big> ====


====== <big>TTS-Modeling</big> ======
====== <big>TTS-Modeling</big> ======
Line 154: Line 154:
* Data Scarcity: For some languages, especially minority or less-resourced ones, there is a scarcity of quality data to train robust models, which hinders the development of universal multi-lingual TTS systems.
* Data Scarcity: For some languages, especially minority or less-resourced ones, there is a scarcity of quality data to train robust models, which hinders the development of universal multi-lingual TTS systems.
* Accent and Dialect Preservation: Preserving native accents and dialects while ensuring clarity and naturalness in synthesized speech across different languages is a complex task.
* Accent and Dialect Preservation: Preserving native accents and dialects while ensuring clarity and naturalness in synthesized speech across different languages is a complex task.
== References ==
<references />


== LLM Review ==
== LLM Review ==
We asked ChatGPT to check the article for spelling and stylistic errors and, if necessary, rewrite certain phrases in a more scientific style, using the prompt "Act as a professor of speech science and review this wiki page, offering advice on how to make it coherent, logical and highlight area fro improvement". We explained every acronyms that ChatGPT suggested to us to do, and unified our reference formant, made the text more coherent, informative. Some of the advice we didn't pick up, such as Structural Improvements, we don't think it is wise to seperate "Key Innovations" and "Articulatory Features-Based TTS",  The page now features a second version verified by ChatGPT. As a result, we received the following review:
We asked ChatGPT to check the article for spelling and stylistic errors and, if necessary, rewrite certain phrases in a more scientific style, using the prompt "Act as a professor of speech science and review this wiki page, offering advice on how to make it coherent, logical and highlight area for improvement". We explained every acronyms that ChatGPT suggested to us to do, and unified our reference formant, made the text more coherent, informative. Some of the advice we didn't pick up, such as Structural Improvements, we don't think it is wise to seperate "Key Innovations" and "Articulatory Features-Based TTS",  The page now features a second version verified by ChatGPT. As a result, we received the following review:


The provided wiki page on "Advancements in Neural Network-Based TTS (2000s)" contains a lot of valuable information about the history, innovations, impact, and future research in the field of neural network-based text-to-speech (TTS). Overall, the content is informative, but there are a few areas where improvements can be made to enhance its clarity and organization. Here are some suggestions:
The provided wiki page on "Advancements in Neural Network-Based TTS (2000s)" contains a lot of valuable information about the history, innovations, impact, and future research in the field of neural network-based text-to-speech (TTS). Overall, the content is informative, but there are a few areas where improvements can be made to enhance its clarity and organization. Here are some suggestions:
Line 178: Line 175:


By addressing these points, the wiki page can become more coherent, informative, and user-friendly for readers interested in the field of neural network-based TTS.
By addressing these points, the wiki page can become more coherent, informative, and user-friendly for readers interested in the field of neural network-based TTS.
== References ==
<references />


== Team Members ==
== Team Members ==

Latest revision as of 16:33, 18 October 2023

Introduction[edit | edit source]

Neural network-based Text-to-Speech (TTS) has a rich history, with roots dating back to early artificial intelligence research. From the 1980s' early explorations of neural networks for speech modeling to the transformative introduction of generative models like WaveNet, this technology has evolved significantly. The quest for natural and expressive speech synthesis led to innovations such as articulatory features-based TTS, DNN-based acoustic modeling system, and vocoder like Tacotron and Transformer. These advancements have enhanced speech quality, prosody modeling, and even ventured into multi-modal synthesis.

Looking ahead, the future of TTS holds promises of efficient synthesis, energy sustainability, and cross-lingual capabilities.

Historical Context[edit | edit source]

The history of neural network-based text-to-speech (TTS) can be traced back to the early days of artificial intelligence research. In the 1980s, researchers began to explore the use of neural networks to model the human speech production process. In the 1990s, Hidden Markov Models were introduced to TTS, which brought significant improvements in speech synthesis. HMM-based systems allowed for better control of speech characteristics and were widely adopted for several years.

In early 2000s, researchers started exploring the use of deep neural networks (DNNs) for speech synthesis. However, it wasn’t until the introduction of generative adversarial networks (GANs) and autoregressive models that the quality of synthesized speech improved significantly. In recent years, the development of deep learning and artificial intelligence has led to a surge in research on neural network-based TTS.[1]

One of the key breakthroughs in neural network-based TTS came in 2006 with the introduction of the WaveNet model by Google AI. WaveNet was the first neural network-based TTS system to generate high-quality speech waveforms directly from text, without the need for intermediate representations such as phonemes or mel spectrograms.[2] This led to a significant improvement in the naturalness and expressiveness of synthesized speech.

Key Innovations[edit | edit source]

Deep Neural Network-Based Text-to-Speech (TTS) has undergone a remarkable transformation whose evolution marks a shift from traditional rule-based and statistical methods towards neural network-driven solutions. Researchers have continually improved the quality and efficiency of TTS systems during this period. In the 2000s, DNN-based TTS models began to gain prominence, paving the way for more natural and expressive speech synthesis. Commencing with the introduction of foundational models, we shall elucidate the subsequent advancements in Text-to-Speech (TTS) technology. In this paragrah, some working mechanism are introduced to readers.[3]

Articulatory Features-Based TTS[edit | edit source]

Articulatory feature-based Text-to-Speech (TTS) is a concept that involves using articulatory features, which represent the movements and positions of the speech articulators (such as the tongue, lips, and jaw), as the basis for synthesizing speech. This approach aims to capture the detailed articulatory information present in the speech signal, allowing for more natural and expressive speech synthesis.[4]

Innovation:

  • Utilization of Biophysical Phonetics: By incorporating articulatory models and other biophysical phonetic information, this approach enhances the quality and naturalness of speech synthesis.
  • Enhanced Speech Quality: Articulatory Features-Based TTS improves the quality of synthesized speech by considering articulatory features, making it more closely resemble natural human speech.
  • Addressing Shortcomings of Traditional TTS: This method aims to compensate for the limitations of traditional TTS systems, particularly in terms of naturalness and quality of speech.
  • Improved Control Capabilities: Articulatory Features-Based TTS offers enhanced control over speech synthesis, enabling users to adjust parameters such as pitch, speed, and other characteristics.
  • Data-Driven Learning: This approach leans towards data-driven learning, reducing reliance on manual rules and models for speech synthesis.
  • These innovations have the potential to enhance the performance of speech synthesis systems, bringing them closer to natural human speech while providing greater control over the synthesized output.[4]

DNN-Based Acoustic TTS-Modeling and Vocoder[edit | edit source]

TTS-Modeling[edit | edit source]

Tacotron: Tacotron is an end-to-end generative Text-to-Speech (TTS) model that directly synthesizes speech from input characters, utilizing a sequence-to-sequence (seq2seq) architecture with attention. It avoids the need for phoneme-level alignment.[5]

Innovation:

  • End-to-End Generative Model: Tacotron is an end-to-end generative model that synthesizes speech directly from characters, eliminating the need for intermediate linguistic features or acoustic models.
  • Sequence-to-Sequence Model with Attention: It is based on a sequence-to-sequence model with attention, enabling high-accuracy and natural speech generation.
  • No Phoneme-Level Alignment Required: Tacotron doesn't necessitate phoneme-level alignment, simplifying scalability to large data with transcripts.
  • Faster Frame-Level Generation: Tacotron generates speech at the frame level, making it considerably faster than sample-level autoregressive methods like WaveNet.
  • High Subjective Mean Opinion Score (MOS): Tacotron attains a 3.82 out of 5 on the subjective mean opinion score, surpassing a production parametric system in terms of naturalness, particularly for US English.

Tacotron 2: Tacotron 2 is a neural network architecture for direct text-to-speech synthesis. It consists of two core elements: a feature prediction network and a modified WaveNet vocoder.[6]

Innovation:

  • Compact Acoustic Intermediate Representation: Tacotron 2 utilizes mel spectrograms, providing a streamlined representation of speech features and reducing WaveNet's architectural complexity.
  • Modified WaveNet Vocoder: Tacotron 2 adapts WaveNet architecture to convert mel spectrograms into time-domain waveform samples, achieving audio quality akin to human speech.
  • Integration of Tacotron and WaveNet: Combining Tacotron-style prosody modeling and WaveNet vocoder, Tacotron 2 delivers state-of-the-art sound quality in a unified, neural approach to speech synthesis.

FastSpeech: FastSpeech is a neural text-to-speech (TTS) system that addresses the challenges of slow inference speed, lack of robustness, and lack of controllability in traditional TTS models. It uses a feed-forward network based on Transformer to generate mel-spectrograms in parallel, allowing for faster synthesis.[7]

Innovation:

  • Multi-head attention mechanism: Enhances long-range dependency modeling and parallelization by allowing simultaneous attention to different parts of the input sequence.
  • Positional encoding: Provides positional information to input sequence elements, aiding in distinguishing elements with identical values.
  • Layer normalization: Improves training stability by normalizing inputs to each network layer.
  • Stacked self-attention layers: Enables the network to learn multiple representation levels of the input sequence, enhancing output quality.
  • No recurrence or convolution: Unlike traditional architectures, the Transformer network omits recurrent connections and convolutions, resulting in improved efficiency and parallelizability.[7]

Transformer: The Transformer is a novel neural network architecture introduced by Vaswani et al. in 2017. It stands out for its exclusive reliance on attention mechanisms, foregoing recurrent connections and convolutions. In natural language processing, particularly neural machine translation, it has achieved remarkable success. The Transformer comprises an encoder and a decoder, both structured with stacks of identity blocks. It employs multi-head self-attention to model dependencies between input and output sequences.[8]

Features:

  • Multi-Head Attention Efficiency: In Transformer, multi-head attention in Tacotron2 enhances training efficiency, constructing hidden states in the encoder and decoder concurrently, speeding up training by 4.25 times and effectively addressing long-range dependencies.
  • WaveNet Vocoder's Role: Within the Transformer TTS network, WaveNet vocoder synthesizes high-quality audio from mel spectrograms, closely resembling human recordings on specific datasets.
  • Transformer Architecture: The Transformer, introduced by Vaswani et al. in 2017, is a unique neural network architecture solely based on attention mechanisms. It excels in natural language processing tasks like neural machine translation, comprising encoder and decoder stacks of identity blocks and utilizing multi-head self-attention to model input-output dependencies.[8]
Vocoder[edit | edit source]

WaveNet: WaveNet serves as a vital component in Tacotron 2 for speech synthesis. It transforms mel-scale spectrograms, predicted by the feature prediction network, into time-domain waveform samples, resulting in high-quality audio waveforms.In this architecture, WaveNet is adapted to function as a vocoder. It takes the predicted mel spectrograms and uses dilated convolutional layers organized into dilation cycles.[6]

Innovation:

  • Dilated Causal Convolutions: Utilizes dilated causal convolutions to exponentially expand the receptive field with the number of layers.
  • Gated Activation Units: Incorporates gated activation units to regulate information flow within the network.
  • Skip Connections: Employs skip connections for the network to learn residual functions.
  • Softmax Output Layer: Utilizes a softmax output layer to model the probability distribution over the next audio sample.
  • Hierarchical Structure: Adopts a hierarchical structure to model audio at multiple scales.[6]


Parallel WaveGAN: Parallel WaveGAN is a waveform generator utilizing GANs. It employs non-autoregressive WaveNet with multi-resolution spectrogram and adversarial loss functions for high-quality speech waveform synthesis. Unlike traditional models, it doesn't require complex density distillation, offering faster, efficient, small-footprint, and competitive waveform generation suitable for real-time applications.

Innovation:

  • Non-Autoregressive WaveNet Generator: Utilizes a non-autoregressive WaveNet generator for faster and more efficient speech synthesis compared to traditional autoregressive models.
  • Multi-Resolution Spectrogram and Adversarial Loss Functions: Trains the generator using multi-resolution spectrogram and adversarial loss functions to capture realistic speech waveforms' time-frequency distribution.
  • No Density Distillation Required: Simplifies the training process by eliminating the need for density distillation, enhancing overall model efficiency.
  • High-Fidelity Speech Generation with Small Model: Achieves high-fidelity speech synthesis with a compact model, suitable for real-time applications.
  • Faster-Than-Real-Time Inference Speed: Provides inference speeds faster than real-time, making it ideal for real-time applications.
  • Competitive Performance: Achieves competitive performance compared to other waveform generation models, ensuring high-quality speech synthesis.[9]

Impact[edit | edit source]

The evolution of neural network models have ushered in transformative impacts across various facets of speech synthesis, notably enhancing the quality, expressiveness, and versatility of synthesized speech.

Enhanced Quality of Speech Synthesis[edit | edit source]

Neural network-based vocoders, such as WaveNet, have significantly improved the quality of synthesized speech, providing more natural and expressive voice outputs. This has been pivotal in reducing the robotic tones often occurs in earlier TTS systems. Using neural network techniques like Tacotron 2 and WaveNet, we can process transcript-free noisy speech datasets in more precise way. Based on it we can create a model capable of generating audio in speakers' voices that not present in the original data.

Prosody Modeling[edit | edit source]

The emergence of neural network models in speech synthesis has dramatically influenced prosody modeling, which involves the prediction and generation of prosodic features like pitch, duration, and energy. It is crucial for producing speech that sounds rhythmic and sentimental. It enables the synthesis of expressive and emotional speech by learning and generating varied prosody, which is crucial for conveying different emotions and speaking styles. Neural network based end-to-end text to speech can facilitate the development of controllable speech synthesis systems where prosody can be manipulated to generate speech with desired pitch, stress, and rhythm.[7]

End-to-End Systems[edit | edit source]

Neural network-based vocoders on End-to-End (E2E) Systems can directly convert text to speech without requiring intermediate representations. It enables systems to learn complex mappings from input to output and often simplifying the traditional multi-stage processing pipeline. Tacotron, an end-to-end generative text-to-speech model can achieve speech synthesis directly from characters, which significantly reduces requirements of expertise of acoustic and many other domain.[5]This has also enabled the creation of speech that can be more dynamically adjusted to various contexts and emotional tones, enhancing applications like virtual assistants and conversational agents.

Generative Modeling[edit | edit source]

Neural networks have enabled the development of generative models that can produce high-quality, natural-sounding speech, improving upon traditional concatenative and parametric methods. It can generate speech with varied emotional content, and can be trained to mimic different speakers, accents, and styles, providing versatility in speech synthesis applications. Generative models have advantages in situations with limited training data to synthesize speech, enabling the creation of voices for speakers with limited available recordings.[5]

Future research[edit | edit source]

Multi-Modal Speech Synthesis[edit | edit source]

Multi-modal speech synthesis refers to the generation of synthetic speech that is not only audible but also visually coherent with facial movements, as we mentioned before in Key Innovations: articulatory features-based TTS. Neural network models, especially generative models like Generative Adversarial Networks (GANs), have been pivotal in synthesizing realistic visual representations (like lip movements) corresponding to synthesized or real speech.[10]

Advantages:

  • Enhanced User Experience: Multi-modal synthesis provides a richer and more immersive user experience by aligning visual cues with synthesized speech.
  • Accessibility: It can enhance communication accessibility, especially for individuals with hearing impairments, by providing visual speech cues.
  • Realistic Virtual Interactions: It enables the creation of realistic virtual characters or digital humans for applications in virtual reality, gaming, and online communication.

Challenges:

  • Lip Synchronization: Ensuring that the synthesized speech is perfectly synchronized with the lip movements to avoid uncanny valley experiences.
  • Expressiveness: Maintaining natural facial expressions and emotions while ensuring lip synchronization can be complex.
  • Data Requirements: Acquiring high-quality, synchronized audio-visual data for training models can be challenging and resource-intensive.
  • Computational Complexity: Managing and processing multiple modalities (audio and visual) requires significant computational resources and optimized algorithms.

Efficient speech synthesis[edit | edit source]

Achieving high-quality speech synthesis propels us towards the pivotal task of efficient synthesis, which encompasses minimizing the costs associated with speech synthesis, such as data collection, labeling, and TTS model training and serving.

Modern neural TTS systems, while capable of synthesizing exquisite speech, typically utilize substantial neural networks, often inhibiting applications in resource-constrained devices like mobiles and IoT due to their extensive memory and power demands. Thus, crafting models that are both compact and lightweight, ensuring reduced memory usage, power consumption, and latency, becomes imperative for such applications.

Moreover, the energy-intensive and carbon-emitting nature of training and serving top-tier TTS models necessitates enhancements in energy efficiency, such as diminishing the FLOPs in TTS training and inference, to broaden accessibility to advanced TTS technologies while concurrently mitigating environmental impact.

Challenges:

  • Balancing Quality and Efficiency: Crafting models that are lightweight yet do not compromise on the quality of speech synthesis.
  • Adaptability: Ensuring that efficient models can adapt to various speakers, emotions, and styles with limited resources.
  • Energy-Efficient Training: Developing training methodologies that require less computational power without sacrificing the learning capability of the models.
  • Low-Resource Adaptation: Ensuring the models can perform optimally even in environments with restricted computational and memory resources.
  • Environmental Sustainability: Aligning the development and usage of TTS technologies with environmental sustainability goals, ensuring that advancements do not exacerbate carbon emissions.

Cross-Lingual and Multi-Lingual Speech Synthesis[edit | edit source]

Cross-lingual and multi-lingual speech synthesis in the realm of Neural Network-based Text-to-Speech (TTS) systems is an intriguing and complex domain, aiming to generate synthesized speech across various languages seamlessly. This area is particularly vital for creating TTS systems that can cater to a global audience, ensuring that technology is accessible and usable across linguistic boundaries.

Firstly, envisioning a future where a single TTS model seamlessly generates speech across multiple languages, the development of a unified phonetic representation becomes imperative. This representation would not only encapsulate the phonetic intricacies of various languages but also serve as a linchpin, enabling the TTS system to navigate through the phonetic landscapes of different languages with finesse.[11]

Moreover, the exploration and advancement of transfer learning techniques hold the potential to bridge the gap between data-rich and data-scarce languages. By harnessing knowledge from languages with abundant data, the technology can be finessed to enhance speech synthesis in languages that are traditionally data-limited, thereby broadening the linguistic horizons of the TTS system.[12]

Simultaneously, the future beckons a deeper dive into adaptive prosody modeling, where the system would dynamically modulate the prosodic elements of synthesized speech to align with the specific contours of the target language. This ensures that the speech is not only linguistically accurate but also rhythmically and melodically congruent with the natural prosody of the language.

Furthermore, embedding cultural and emotional nuances in synthesized speech emerges as a pivotal frontier. The future TTS system would not merely be a linguistic translator but a cultural and emotional interpreter, ensuring that the synthesized speech resonates authentically, both linguistically and emotionally, across varied cultural contexts.

In synthesizing these pathways—crafting a unified phonetic representation, leveraging transfer learning, delving into adaptive prosody modeling, and embedding cultural and emotional nuances—the future of TTS technology is sculpted. A future where the technology is not just a tool for linguistic translation but a conduit for authentic, emotionally resonant, and culturally rich communication across a tapestry of languages and cultures.

Challenges:

  • Phonetic and Prosodic Variations: Different languages have distinct phonetic and prosodic characteristics. Modeling these variations accurately to generate natural-sounding speech in multiple languages is challenging.
  • Data Scarcity: For some languages, especially minority or less-resourced ones, there is a scarcity of quality data to train robust models, which hinders the development of universal multi-lingual TTS systems.
  • Accent and Dialect Preservation: Preserving native accents and dialects while ensuring clarity and naturalness in synthesized speech across different languages is a complex task.

LLM Review[edit | edit source]

We asked ChatGPT to check the article for spelling and stylistic errors and, if necessary, rewrite certain phrases in a more scientific style, using the prompt "Act as a professor of speech science and review this wiki page, offering advice on how to make it coherent, logical and highlight area for improvement". We explained every acronyms that ChatGPT suggested to us to do, and unified our reference formant, made the text more coherent, informative. Some of the advice we didn't pick up, such as Structural Improvements, we don't think it is wise to seperate "Key Innovations" and "Articulatory Features-Based TTS",  The page now features a second version verified by ChatGPT. As a result, we received the following review:

The provided wiki page on "Advancements in Neural Network-Based TTS (2000s)" contains a lot of valuable information about the history, innovations, impact, and future research in the field of neural network-based text-to-speech (TTS). Overall, the content is informative, but there are a few areas where improvements can be made to enhance its clarity and organization. Here are some suggestions:

  • Structural Improvements: Consider breaking down the content into subsections with clear headings for better navigation. For example, separate the "Key Innovations" section from the "Articulatory Features-Based TTS" section to improve the page's overall structure.
  • Consistent Terminology: Ensure consistent use of terminology and acronyms throughout the page. For example, it would be beneficial to explain or provide links for less common acronyms such as "MOS."
  • Citations: Provide clear citations for the sources of the information, especially when referencing research papers and specific models. This adds credibility to the content.
  • Articulatory Features-Based TTS: Expand the section on "Articulatory Features-Based TTS" to include a brief explanation of what articulatory features are and why they are important in speech synthesis. This will help readers who may not be familiar with the concept.
  • Visual Aids: Consider including visual aids like diagrams or flowcharts to help illustrate complex concepts, such as the structure of the WaveNet model or the operation of Tacotron.
  • Prosody Modeling: Expand on the importance of prosody modeling, providing examples of how it impacts speech synthesis. Explain how neural network-based models have improved prosody modeling.
  • End-to-End Systems: Highlight the advantages and disadvantages of end-to-end systems more explicitly. Discuss cases where they excel and where they may have limitations.
  • Generative Modeling: Provide more context on the role of generative models in speech synthesis and how they compare to traditional methods like concatenative and parametric synthesis.
  • Multi-Modal Speech Synthesis: Include examples or use cases of multi-modal speech synthesis to make it more tangible for readers. Explain how it benefits individuals with hearing impairments, for instance.
  • Efficient Speech Synthesis: Elaborate on the challenges of achieving energy-efficient training and how it relates to the use of neural network models in TTS.
  • Cross-Lingual and Multi-Lingual Speech Synthesis: Offer more insights into the challenges of handling different languages, such as phonetic and syntactic variations, without interfering with each other.
  • Conclusion and Future Research: Summarize the main takeaways from the article and reiterate the importance of neural network-based TTS advancements in the broader context of AI and speech technology.
  • References: Include a section for references, listing the sources and publications that were cited throughout the article.

By addressing these points, the wiki page can become more coherent, informative, and user-friendly for readers interested in the field of neural network-based TTS.

References[edit | edit source]

  1. Xu Tan∗, Tao Qin, Frank Soong, Tie-Yan Liu. "A Survey on Neural Speech Synthesis" arXiv:2106.15561 (2021)
  2. Dario Amodei, Dario, Aidan N. Gomez, et al. "WaveNet: A Generative Model for Raw Audio." arXiv preprint arXiv:1609.03499 (2016).
  3. G. Hinton et al., "Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups," in IEEE Signal Processing Magazine, vol. 29, no. 6, pp. 82-97, Nov. 2012, doi: 10.1109/MSP.2012.2205597.
  4. 4.0 4.1 Singampalli, V. D. (2010). Statistical identification of articulatory roles in speech production (Order No. 10131268). Available from ProQuest Dissertations & Theses A&I. (1810640121). Retrieved from http://server.proxy-ub.rug.nl/login?url=https://www.proquest.com/dissertations-theses/statistical-identification-articulatory-roles/docview/1810640121/se-2
  5. 5.0 5.1 5.2 Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). "Tacotron: Towards End-to-End Speech Synthesis." In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (pp. 1946-1954).
  6. 6.0 6.1 6.2 Wang, Y., Skerry-Ryan, R. J., Stanton, D., Wu, Y., Weiss, R. J., Jaitly, N., ... & van den Oord, A. (2017). Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions. arXiv preprint arXiv:1712.05884.
  7. 7.0 7.1 7.2 Ren, J., Xu, L., Zhang, Z., Yang, T., Lai, J., Lu, Z., & Dai, L. R. (2019). FastSpeech: Fast, Robust and Controllable Text to Speech. arXiv preprint arXiv:1905.09263.
  8. 8.0 8.1 Naihan Li, Shujie Liu, Yanqing Liu, Sheng Zhao, and Ming Liu. 2019. Neural speech synthesis with transformer network. In Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence and Thirty-First Innovative Applications of Artificial Intelligence Conference and Ninth AAAI Symposium on Educational Advances in Artificial Intelligence (AAAI'19/IAAI'19/EAAI'19). AAAI Press, Article 823, 6706–6713. https://doi.org/10.1609/aaai.v33i01.33016706
  9. Yamamoto, R., Inoue, K., Portnoff, M., Tan, X., Inoue, S., Yamamoto, H., ... & Watanabe, S. (2020). Parallel WaveGAN: A fast waveform generation model based on generative adversarial networks with multi-resolution spectrogram. arXiv preprint arXiv:1910.11480.
  10. Hang Zhou, Yu Liu, Ziwei Liu, Ping Luo, Xiaogang Wang. Talking Face Generation by Adversarially Disentangled Audio-Visual Representation
  11. Sungjin Lee, Qi Zhu, Ryuichi Takanobu, Zheng Zhang, Yaoqin Zhang, Xiang Li, Jinchao Li, Baolin Peng, Xiujun Li, Minlie Huang, Jianfeng Gao. ConvLab: Multi-Domain End-to-End Dialog System Platform
  12. Ye Jia, Yu Zhang, Ron J. Weiss, Quan Wang, Jonathan Shen, Fei Ren, Zhifeng Chen, Patrick Nguyen, Ruoming Pang, Ignacio Lopez Moreno. Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis

Team Members[edit | edit source]

Qing Li

Lifan Qu

Yi Lei