Tacotron: Towards End-to-End Speech Synthesis
The DOI matches its Crossref record; 25 of the 25 references it cites were found in a registry or an index.
Abstract
A text-to-speech synthesis system typically consists of multiple stages, such as a text analysis frontend, an acoustic model and an audio synthesis module.Building these components often requires extensive domain expertise and may contain brittle design choices.In this paper, we present Tacotron, an end-to-end generative text-to-speech model that synthesizes speech directly from characters.Given pairs, the model can be trained completely from scratch with random initialization.We present several key techniques to make the sequence-tosequence framework perform well for this challenging task.Tacotron achieves a 3.82 subjective 5-scale mean opinion score on US English, outperforming a production parametric system in terms of naturalness.In addition, since Tacotron generates speech at the frame level, it's substantially faster than sample-level autoregressive methods.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“The frameworks including their code as well as the datasets used that are available publicly (Sotelo et al., 2017; Wang et al., 2017; Ping et al., 2018; Shen et al., 2018; Vasquez and Lewis, 2019) are more likely to be used for nefarious purposes or research, so they will be more developed.”
From Audio deepfakes: A survey (2023) · cites this paper as Wang et al., 2017
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 2×
“We further pass the processed sequence to a few fixed-width 1-D convolutions, whose outputs are added with the original input sequence via residual connections [17].”
BackgroundAdam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946 · cited 2×
“We use the Adam optimizer (Kingma & Ba, 2015) with learning rate decay, which starts from 0.001 and is reduced to 0.0005, 0.0003, and 0.0001 after 500K, 1M and 2M global steps, respectively.”
MethodInfluentialBatch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Ioffe, S., & Szegedy, C. (2015) · arXiv · doi:10.48550/arxiv.1502.03167 · Cited by 27,818 · cited 2×
“Batch normalization [18] is used for all convolutional layers.”
MethodSequence to Sequence Learning with Neural Networks
Sutskever, I., Vinyals, O., & Le, Q. V. (2014) · arXiv · doi:10.48550/arxiv.1409.3215 · Cited by 17,766 · cited 2×
“In this paper, we propose Tacotron, an end-to-end generative TTS model based on the sequence-to-sequence (seq2seq) (Sutskever et al., 2014) with attention paradigm (Bahdanau et al., 2014).”
MethodSelf-citationNeural Machine Translation by Jointly Learning to Align and Translate
Bahdanau, D., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1409.0473 · Cited by 16,213 · cited 4×
“The backbone of Tacotron is a seq2seq model with attention [7, 14].”
MethodInfluentialEmpirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling
Chung, J., Gulcehre, C., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1412.3555 · Cited by 11,008 · cited 7×
“Second, we compare with a model with the CBHG encoder replaced by a 2-layer residual GRU encoder.”
BackgroundInfluentialTensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
Abadi, M., Agarwal, A., Barham, P., Brevdo, E., Chen, Z., Citro, C., Corrado, G. S., Davis, A., Dean, J., Devin, M., et al. (2016) · arXiv · doi:10.48550/arxiv.1603.04467 · Cited by 9,408 · cited 2×
“We implemented Griffin-Lim in TensorFlow [22] hence it’s also part of the model.”
MethodSelf-citationGoogle's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
Wu, Y., Schuster, M., Chen, Z., Le, Q. V., Norouzi, M., Macherey, W., Krikun, M., Cao, Y., Gao, Q., Macherey, K., et al. (2016) · arXiv · doi:10.48550/arxiv.1609.08144 · Cited by 5,319 · cited 4×
“C L ] 6 A pr or machine translation (Wu et al., 2016), TTS outputs are continuous, and output sequences are usually much longer than those of the input.”
BackgroundMethodInfluentialSelf-citationWaveNet: a generative model for raw audio
Oord, A. V. D., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., Kalchbrenner, N., Senior, A., & Kavukcuoglu, K. (2016) · arXiv · doi:10.48550/arxiv.1609.03499 · Cited by 4,911 · cited 4×
“It could be used to predict alternative targets such as vocoder parameters, or as a WaveNetlike neural vocoder [9, 13, 10] that synthesizes waveform samples directly.”
BackgroundMethodInfluentialListen, attend and spell: A neural network for large vocabulary conversational speech recognition
Chan, W., Jaitly, N., Le, Q., & Vinyals, O. (2016) · 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2016.7472621 · Cited by 2,226 · cited 2×
“Moreover, unlike end-to-end speech recognition [4] or machine translation [5], TTS outputs are continuous, and output sequences are usually much longer than those of the input.”
BackgroundSelf-citation
Cited by
Items 901–1,000 of 1,002
Li, W., Lei, S., Huang, Q., Zhou, Y., Wu, Z., Kang, S., & Meng, H. M. L. (2023) · Interspeech (USB) · doi:10.21437/interspeech.2023-1754 · Cited by 11
MuLanTTS: The Microsoft Speech Synthesis System for Blizzard Challenge 2023
Xu, Z., Zhang, S., Wang, X., Zhang, J., Wei, W., He, L., & Zhao, S. (2023) · doi:10.21437/blizzard.2023-5 · Cited by 9
Xuan, H., Liu, X., Tan, Y., & Zhou, Q. (2023) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3581783.3613841 · Cited by 12
End-to-End Mongolian Text-to-Speech System
Li, J., Zhang, H., Liu, R., Zhang, X., & Bao, F. (2018) · International Symposium on Chinese Spoken Language Processing (ISCSLP) · doi:10.1109/iscslp.2018.8706263 · Cited by 8
“[14].”
BackgroundMethodInfluentialTits, N. (2019) · International Conference on Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW) · doi:10.1109/aciiw.2019.8925241 · Cited by 12
“systems ability to synthesize speech with high quality and naturalness such as Wavenet [1], Tacotron [5], Deep Convolutional TTS [7] (DCTTS), WaveRNN [8], Char2Wav [9] and Deep Voice [10].”
BackgroundMethodInfluentialJaiswal, J., Chaubey, A., Bhimavarapu, S. K. R., Kashyap, S., Kumar, P., Raman, B., & Roy, P. P. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-41299-9_45 · Cited by 8
MAKEDONKA: Applied Deep Learning Model for Text-to-Speech Synthesis in Macedonian Language
Mishev, K., Ристовска, А. К., Trajanov, D., Eftimov, T., & Simjanoska, M. (2020) · Applied Sciences · doi:10.3390/app10196882 · Cited by 8
“One of the most famous and widely used is Tacotron, which is a seq2seq model [16], and its successor Tacotron2 [17,18].”
MethodCardenas, I. S., Lenhoff, C., Park, M., Xu, T. Y., Lin, X., Paladugula, P. K., & Kim, J.-H. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-68452-5_42 · Cited by 9
Joint Alignment Learning-Attention Based Model for Grapheme-to-Phoneme Conversion
Wang, Y., Bao, F., Zhang, H., & Gao, G. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9413679 · Cited by 8
Self-citationImproving Performance of Seen and Unseen Speech Style Transfer in End-to-End Neural TTS
An, X., Soong, F. K., & Xie, L. (2021) · Interspeech (USB) · doi:10.21437/interspeech.2021-1407 · Cited by 10
Towards a Versatile Intelligent Conversational Agent as Personal Assistant for Migrants
Wanner, L., Klusch, M., Mavropoulos, A., Jamin, E., Puchades, V. M., Casamayor, G., Černocký, J., Davey, S., Domínguez, M., Egorova, E., et al. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-85739-4_26 · Cited by 8
FastTacotron: A Fast, Robust and Controllable Method for Speech Synthesis
Sang, D. V., & Thu, L. X. (2021) · International Conference on Multimedia Analysis and Pattern Recognition (MAPR) · doi:10.1109/mapr53640.2021.9585267 · Cited by 8
Naderi, N., Nasersharif, B., & Nikoofard, A. (2021) · Multimedia Tools and Applications · doi:10.1007/s11042-021-11719-w · Cited by 11
“In the decoder part, each time frame of Mel spectrogram is generated from encoded output values and previous output of the decoder at each time step (auto-regressive architecture) [37].”
MethodNishimura, Y., Saito, Y. C., Takamichi, S., Tachibana, K., & Saruwatari, H. (2022) · Interspeech (USB) · doi:10.21437/interspeech.2022-403 · Cited by 11
AdaVITS: Tiny VITS for Low Computing Resource Speaker Adaptation
Song, K., Xue, H., Wang, X., Cong, J., Zhang, Y., Xie, L., Yang, B., Zhang, X., & Su, D. (2022) · International Symposium on Chinese Spoken Language Processing (ISCSLP) · doi:10.1109/iscslp57327.2022.10037585 · Cited by 12
“To create human-like natural speech, modern neural networkbased text-to-speech (TTS) models are usually large and contain mass of parameters [2, 3, 4, 5].”
BackgroundEmotional Speech Synthesis for Multi-Speaker Emotional Dataset Using WaveNet Vocoder
Choi, H., Park, S., Park, J., & Hahn, M. (2019) · International Conference on Consumer Electronics · doi:10.1109/icce.2019.8661919 · Cited by 8
“The mel spectrograms are converted to waveforms, using post network [5], [6] followed by Griffin-Lim [7] or WaveNet vocoder [8].”
MethodGraphPB: Graphical Representations of Prosody Boundary in Speech Synthesis
Sun, A., Wang, J., Cheng, N., Peng, H., Zeng, Z., Kong, L., & Xiao, J. Z. (2021) · IEEE Spoken Language Technology Workshop (SLT) · doi:10.1109/slt48900.2021.9383530 · Cited by 9
“The prosody boundary information [18] is tried to be embedded into the Tacotron [19, 20] model by [21] through adding a context encoder analysing the context information.”
MethodInfluentialCaption Generation of Robot Behaviors Based on Unsupervised Learning of Action Segments
Yoshino, K., Wakimoto, K., Nishimura, Y., & Nakamura, S. (2020) · Lecture notes in electrical engineering · doi:10.1007/978-981-15-8395-7_17 · Cited by 8
Keskin, G., Lee, T., Stephenson, C., & Elibol, O. H. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.12531 · Cited by 8
“Tacotron-based TTS models (Wang et al., 2017; Shen et al., 2018) have been used to generate synthetic speech to improve ASR performance.”
MethodLearning Singing From SpeechPreprint
Zhang, L., Yu, C., Lu, H., Weng, C., Wu, Y., Xie, X., Li, Z., & Yu, D. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1912.10128 · Cited by 8
“The output from the decoder network y ′ 1:T is passed through a post-CBHG [13] to improve the quality of predicted mel-spectrogram as”
MethodSelf-citationPerquin, A., Cooper, E., & Yamagishi, J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.10694 · Cited by 8
“Such end-to-end systems include Tacotron [4], which has been shown to outperform traditional approaches when paired with a high-quality vocoder [5].”
BackgroundItôTTS and ItôWave: Linear Stochastic Differential Equation Is All You Need For Audio GenerationPreprint
Wu, S., & Shi, Z. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2105.07583 · Cited by 14
“To name a few, for example, in the field of TTS, AR-type models are generally implemented through the framework of encoder and decoder, and the duration of text is predicted by an attention module, such as Tacotron [36] and Tacotron 2 [28], DeepVoice 2 [7] and 3 [22].”
MethodInfluentialBeyond $L_p$ clipping: Equalization-based Psychoacoustic Attacks against ASRsPreprint
Abdullah, H., Rahman, M. S., Peeters, C., Gibson, C., García, W., Bindschaedler, V., Shrimpton, T., & Traynor, P. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.13250 · Cited by 8
“The most fundamental and widely used Shen et al. (2018); Wang et al. (2017) technique for spectrogram inversion is the Griffin-Lim algorithm Griffin and Lim (1984).”
MethodAudio representations for deep learning in sound synthesis: A reviewPreprint
Natsiou, A., & O’Leary, S. T. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2201.02490 · Cited by 9
“This type of spectrograms has been used in many by a variety of papers [18] [19] [20].”
BackgroundAttention Forcing for Sequence-to-sequence Model TrainingPreprint
Dou, Q., Lu, Y., Efiong, J., & Gales, M. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1909.12289 · Cited by 8
“The model structure and configuration are the same as described in Wang et al. (2017), except that: 1) the decoder target is vocoder features; 2) the attention mechanism is the hybrid (content-based + location-based) attention (Chorowski et al., 2015); 3) each decoding step predicts 5 vocoder…”
MethodResultInfluentialDeep generative models for musical audio synthesisPreprint
Huzaifah, M., & Wyse, L. L. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.06426 · Cited by 8
“In a similar spirit to SING, Tacotron [Wang et al., 2017a] also implements RNN-based networks to sequentially predict mel spectrograms for speech, contingent upon embeddings for speaker, text, and reference spectrograms.”
BackgroundMethodSynthetic Tabular Data Generation: A Comparative Survey for Modern TechniquesPreprint
Challagundla, R., Dorodchi, M., Wang, P., & Lee, M. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.11590 · Cited by 9
Survey: Transformer-based Models in Data Modality ConversionPreprint
Rashno, E., Eskandari, A., Anand, A., & Zulkernine, F. H. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2408.04723 · Cited by 11
“WavLM [26] is a pre-trained model for speech processing, trained on 94,000 hours of audio using self-supervised MOS score is a rating humans give to judge audio quality [171].”
MethodBailo, P., Sirignano, A., Nittari, G., Visconti, G., Pesel, G., Spasari, T., & Ricci, G. (2026) · Sci · doi:10.3390/sci8030054 · Cited by 8
“The rapid proliferation of deepfakes is paralleled by voice-cloning technologies that re-produce human speech with high fidelity using deep neural architectures such as Tacotron 2, WaveNet, and VALL-E, enabling capture of timbre, prosody, and emotional tone from only a few seconds of audio [19,20].”
Mongolian emotional speech synthesis based on transfer learning and emotional embedding
Huang, A., Bao, F., Gao, G., Yu, S., & Liu, R. (2021) · International Conference on Asian Language Processing · doi:10.1109/ialp54817.2021.9675192 · Cited by 7
Reconstructing Dual Learning for Neural Voice Conversion Using Relatively Few Samples
Sun, A., Wang, J., Cheng, N., Tantrawenith, M., Wu, Z., Meng, H. M. L., Xiao, E., & Xiao, J. Z. (2021) · IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU) · doi:10.1109/asru51503.2021.9687965 · Cited by 7
Text to speech using Mel-Spectrogram with deep learning algorithms
Karim, A. A., & Saleh, S. M. (2022) · Periodicals of Engineering and Natural Sciences (PEN) · doi:10.21533/pen.v10.i3.671 · Cited by 7
Semi-Supervised Learning Based on Reference Model for Low-resource TTS
Zhang, X., Wang, J., Cheng, N., & Xiao, J. Z. (2022) · International Conference on Mobility, Sensing and Networking (MSN) · doi:10.1109/msn57253.2022.00156 · Cited by 8
“Tacotron [6] could directly generates waveform signals from input text.”
BackgroundPersonalized Lightweight Text-to-Speech: Voice Cloning with Adaptive Structured Pruning
Huang, S.-F., Chen, C.-P., Chen, Z., Tsai, Y.-P., & Lee, H.-Y. (2023) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49357.2023.10097178 · Cited by 10
“To train a high-quality single-speaker end-to-end TTS [1, 2, 3, 4, 5], hours of single-speaker speech recordings [6] and large, specially designed models are required.”
BackgroundSelf-citationMulti speaker text-to-speech synthesis using generalized end-to-end loss function
Nazir, O., Malik, A., Singh, S., & Pathan, A. K. (2024) · Multimedia Tools and Applications · doi:10.1007/s11042-024-18121-2 · Cited by 8
“Tacotron [4], an end-to-end TTS using the Griffin-Lim algorithm [5] as its vocoder leveraged the process to a certain degree.”
Making Social Platforms Accessible: Emotion-Aware Speech Generation with Integrated Text Analysis
De, S., Bostan, I., & Sastry, N. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-78554-2_7 · Cited by 7
Ahn, Y., Chae, J.-W., & Shin, J. W. (2025) · IEEE Signal Processing Letters · doi:10.1109/lsp.2025.3537949 · Cited by 8
“The performance of the TTS has been sig-nificantly improved by deep learning-based approaches [5], [5], [6], [6], [7], [8], [9], [10], and the tasks of the TTS have expanded to the synthesis of the speech from unseen speakers [11], [12], [13], multiple languages [14], [15], [16], and with emotions…”
Real-Time Voice Cloning: Artificial Intelligence to Clone and Generate Human Voice
Patel, A., Madnani, H., Tripathi, S., Sharma, P., & Shukla, V. K. (2025) · Lecture notes in electrical engineering · doi:10.1007/978-981-97-8193-5_29 · Cited by 7
JAM-Flow: Joint Audio-Motion Synthesis with Flow Matching
Kwon, M., Shin, J., Jeong, J., Park, J., & Uh, Y. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-37553-7_20 · Cited by 8
Towards End-to-End Raw Audio Music Synthesis
Eppe, M., Alpay, T., & Wermter, S. (2018) · Lecture notes in computer science · doi:10.1007/978-3-030-01424-7_14 · Cited by 7
“It has inspired the Tacotron text-to-speech framework which provides expressive results in speech synthesis [18].”
BackgroundMethodMyanmar Text-to-Speech System based on Tacotron-2
Win, Y., & Masada, T. (2020) · International Conference on Information and Communication Technology Convergence (ICTC) · doi:10.1109/ictc49870.2020.9289599 · Cited by 7
“Furthermore, neural network models such as Tacotron [14] and Tacotron-2 [10] obtain a naturalness score as good as human speech by generating mel spectrogram directly from text.”
MethodSanjay, G., Sooraj, K. C., & Mishra, D. (2020) · International Conference on Soft Computing & Machine Intelligence (ISCMI) · doi:10.1109/iscmi51676.2020.9311564 · Cited by 7
Non-Parallel Many-To-Many Voice Conversion by Knowledge Transfer from a Text-To-Speech Model
Yu, X., & Mak, B. K.-W. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9414757 · Cited by 7
“Recently, text-to-speech (TTS) models manage to generate highly natural and intelligible speech [18, 19, 20].”
BackgroundNguyen, H.-K., Jeong, K., Um, S., Hwang, M.-J., Song, E., & Kang, H.-G. (2021) · Interspeech (USB) · doi:10.21437/interspeech.2021-188 · Cited by 7
“Some well-known autoregressive TTS models include Tacotron [1], Tacotron 2 [2], Deep Voice 3 [3], and Transformer TTS [4].”
MethodCross-Lingual Neural Network Speech Synthesis Based on Multiple Embeddings.
Nosek, T., Suzić, S., Pekar, D. J., Obradović, R., Sečujski, M., & Delić, V. (2021) · International Journal of Interactive Multimedia and Artificial Intelligence · doi:10.9781/ijimai.2021.11.005 · Cited by 7
Speech synthesis with face embeddings
Wu, X., Ji, S., Wang, J., & Guo, Y. (2022) · Applied Intelligence · doi:10.1007/s10489-022-03227-7 · Cited by 8
“With the development of deep learning, the TTS systems based on DNN have emerged [30] and WaveNet [31] and Tacotron [32] were proposed and breakthroughs were made in the TTS field.”
MethodInfluentialHow Does a Spontaneously Speaking Conversational Agent Affect User Behavior?
Iizuka, T., & Mori, H. (2022) · IEEE Access · doi:10.1109/access.2022.3214977 · Cited by 9
“Recent studies on speech synthesis [22], [23], [24] have dramatically improved the quality of synthesized speech to the point where it is difficult to distinguish the synthesized speech from a real human voice.”
BackgroundAutolv: Automatic Lecture Video Generator
Wang, W., Song, Y., & Jha, S. K. (2022) · Proceedings - International Conference on Image Processing · doi:10.1109/icip46576.2022.9897436 · Cited by 9
“2, our speech synthesizer is based on Tacotron [1].”
MethodComparing the performance of classic voice-driven assistive systems for dysarthric speech
Zheng, W.-Z., Han, J.-Y., Cheng, H.-L., Chu, W.-C., Chen, K.-C., & Lai, Y.-H. (2022) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2022.104447 · Cited by 7
QSpeech: Low-Qubit Quantum Speech Application Toolkit
Hong, Z., Wang, J., Qu, X., Zhao, C., Wei, T., & Xiao, J. (2022) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn55064.2022.9892496 · Cited by 9
“QTacotron has same the same parameter as as [43].”
BackgroundResultAutotts: End-to-End Text-to-Speech Synthesis Through Differentiable Duration Modeling
Nguyen, B., Cardinaux, F., & Uhlich, S. (2023) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49357.2023.10095431 · Cited by 8
“Many TTS systems are based on autoregressive models such as Tacotron [1,2], Deep Voice [3,4], and Transformer TTS [5].”
MethodAdversarial Text-to-Speech for low-resource languages
Elneima, A. H., & Bińkowski, M. (2022) · Arabic Natural Language Processing Workshop (WANLP) · doi:10.18653/v1/2022.wanlp-1.8 · Cited by 7
AI Enhancements for Linguistic E-Learning System
Liu, J., Li, S., Ren, C., Lyu, Y., Xu, T., Wang, Z., & Chen, W. (2023) · Applied Sciences · doi:10.3390/app131910758 · Cited by 7
Ai, Y., & Ling, Z.-H. (2024) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2024.3385285 · Cited by 7
“…speech generation tasks, such as speech enhancement (SE) [2], [3], [4], bandwidth extension (BWE) [5], [6], [7] and speech synthesis (SS) [8], [9], [10], [11], [12], [13], mainly focus on the prediction of amplitude spectra or amplitude-derived features (e.g., mel spectrograms and mel…”
BackgroundVRAIN-UPV MLLP’s system for the Blizzard Challenge 2021
Pérez-González-de-Martos, A., Sanchís, A., & Juan, A. (2021) · The Blizzard Challenge · doi:10.21437/blizzard.2021-5 · Cited by 7
Cultures of the AI paralinguistic in voice cloning tools
Ada, A. A., Jørgensen, S. H., & Fritsch, J. (2024) · Designing Interactive Systems Conference · doi:10.1145/3656156.3663708 · Cited by 8
“The development of text-to-speech machine learning-based voice cloning gained traction with the release of Tacotron [23].”
BackgroundSpeech-Forensics: Towards Comprehensive Synthetic Speech Dataset Establishment and Analysis
Ji, Z., Lin, C., Wang, H., & Shen, C. (2024) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2024/46 · Cited by 11
Make-A-Voice: Revisiting Voice Large Language Models as Scalable Multilingual and Multitask Learners
Huang, R., Zhang, C., Wang, Y., Yang, D., Tian, J., Ye, Z., Liu, L., Wang, Z., Jiang, Z. K., Chang, X., et al. (2024) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2024.acl-long.589 · Cited by 12
“Text-guided voice synthesis (text-to-speech and singing voice synthesis) typically converts input text into mel-spectrogram (e.g., Tacotron (Wang et al., 2017), FastSpeech (Ren et al., 2019)), which is then transformed to waveform using a separately trained vocoder (Kong et al., 2020; Huang et al., 2021).”
Self-citationAccented Text-to-Speech Synthesis with a Conditional Variational Autoencoder
Melechovský, J., Mehrish, A., Şişman, B., & Herremans, D. (2024) · IEEE Region 10 Conference (TENCON) · doi:10.1109/tencon61640.2024.10902981 · Cited by 9
“In speech synthesis, targeting accent is mostly a domain of Foreign Accent Conversion (FAC) frameworks [11, 12, 13].”
A New End-to-End Long-Time Speech Synthesis System Based on Tacotron2
Liu, R., Yang, J., & Liu, M. (2019) · International Symposium on Signal Processing Systems (SSPS) · doi:10.1145/3364908.3365292 · Cited by 7
“Tacotron2 combines the advantages of WaveNet and Tacotron([4]) to directly output speech by input text without any linguistic knowledge.”
BackgroundXiao, Z., Ling, Z.-H., & Dai, L.-R. (2020) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp40776.2020.9053812 · Cited by 7
“With the development of neural network-based SPSS, endto-end SPSS models have been proposed, such as Tacotron [10, 11], which adopts the sequence-to-sequence framework to combine front-end text analysis, duration modeling and acoustic modeling.”
MethodProsodic Boundary Prediction Model for Vietnamese Text-To-Speech
Trang, N. T. T., Ky, N. H., Rilliard, A. O. B., & d’Alessandro, C. R. (2021) · Interspeech (USB) · doi:10.21437/interspeech.2021-125 · Cited by 9
“In end-to-end TTS systems, such as Tacotron [4][5], Tacotron2 [6], and Wavenet [7], we could use a massive amount of text and audio data pairs to learn the prosodic structure directly during the TTS training process.”
MethodInfluentialData-Efficient Training Strategies for Neural TTS Systems
Prajwal, K. R., & Jawahar, C. V. (2020) · Joint International Conference on Data Science & Management of Data (CODS-COMAD) · doi:10.1145/3430984.3431034 · Cited by 7
“One of the first popularneuraltext-to-speechmodelsintroducedistheTacotron[21] architecture.”
BackgroundInfluentialZero-Shot Text-to-Speech for Text-Based Insertion in Audio Narration
Tang, C., Luo, C., Zhao, Z., Yin, D., Zhao, Y., & Zeng, W. (2021) · Interspeech (USB) · doi:10.21437/interspeech.2021-189 · Cited by 10
GC-TTS: Few-shot Speaker Adaptation with Geometric Constraints
Kim, J., Lee, S., Lee, J.-H., Jung, H.-G., & Lee, S. (2021) · Conference proceedings/Conference proceedings - IEEE International Conference on Systems, Man, and Cybernetics · doi:10.1109/smc52423.2021.9658830 · Cited by 9
“• Fine-tuning is a strong comparable system that finetunes the whole model of Tacotron2 without using two geometric constraints and freezing weights in a similar scheme to [15], [16].”
MethodInfluentialWord-Level Style Control for Expressive, Non-attentive Speech Synthesis
Klapsas, K., Ellinas, N., Sung, J. S., Park, H.-M., & Raptis, S. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-87802-3_31 · Cited by 7
“Since neural text-to-speech models such as Tacotron [22,16] significantly improved the quality and naturalness of speech synthesis systems, there has been an increased interest in approaches that expand on the basic architecture by explicitly modeling the prosody of the synthesized speech.”
BackgroundText-to-Speech with Model Compression on Edge Devices
Koc, W.-W., Chang, Y., Yu, J.-Y., & İk, T.-U. (2021) · Asia-Pacific Network Operations and Management Symposium (APNOMS) · doi:10.23919/apnoms52696.2021.9562651 · Cited by 7
“[5].”
Securing Voice Communications Using Audio Steganography
Phipps, A., Ouazzane, K., & Vassilev, V. (2022) · International Journal of Computer Network and Information Security · doi:10.5815/ijcnis.2022.03.01 · Cited by 12
“In another method demonstrated by Seymore et al. which was more robust to factors such as noise and interference was utilising open-source tools Tacotron [25, 26] and Wavenet [27] and utilising voice samples freely available to the public.”
MethodUnify and Conquer: How Phonetic Feature Representation Affects Polyglot Text-To-Speech (TTS)
Sanchez, A., Falai, A., Zhang, Z., Angelini, O., & Yanagisawa, K. (2022) · Interspeech (USB) · doi:10.21437/interspeech.2022-233 · Cited by 7
Analysis of Guzheng Music Style Transformation Based on Generative Adversarial Networks
Zhang, L. (2022) · Mobile Information Systems · doi:10.1155/2022/4950231 · Cited by 7
“(e whole scheme includes two parts: calculating the difference between different styles of music and calculating the dynamic weight of loss function based on the difference between styles of music [27].”
BackgroundDIA-TTS: Deep-Inherited Attention Based Text-to-Speech Synthesizer
Yu, J., Xu, Z., He, X., Wang, J., Liu, B., Feng, R., Zhu, S., Wang, W., & Li, J. (2022) · Entropy · doi:10.2139/ssrn.4257520 · Cited by 7
Urdu Text-to-Speech Conversion Using Deep Learning
Saba, R., Bilal, M., Ramzan, M., Khan, H. U., & Ilyas, M. (2022) · International Conference on IT and Industrial Technologies (ICIT) · doi:10.1109/icit56493.2022.9989175 · Cited by 7
“Tacotron [4], Tacotron 2 [5], Deep Voice 3 [6], and ClariNet [7] are some of the most popular DNN-based TTS techniques.”
BackgroundTimbre-Reserved Adversarial Attack in Speaker Identification
Wang, Q., Yao, J., Zhang, L., Guo, P., & Xie, L. (2023) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2023.3306714 · Cited by 9
“Speech synthesis, also known as text-to-speech (TTS), takes arbitrary text as input and generates speech as output [34], [35].”
BackgroundBoosting Character-Based Chinese Speech Synthesis via Multi-Task Learning and Dictionary Tutoring
Zou, Y., Dong, L., & Xu, B. (2019) · Interspeech (USB) · doi:10.21437/interspeech.2019-3233 · Cited by 7
“This post processing network is simply a CBHG block in Tacotron [7] followed by a linear projection layer.”
BackgroundMethodSon, S., Jeong, Y., & Lee, B.-K. (2019) · Sensors · doi:10.3390/s19225035 · Cited by 7
“It executes text-to-speech (TTS) using Google’s Tacotron2 [23].”
Zero-Shot Code-Switching ASR and TTS with Multilingual Machine Speech Chain
Nakayama, S., Tjandra, A., Sakti, S., & Nakamura, S. (2019) · IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU) · doi:10.1109/asru46091.2019.9003926 · Cited by 7
SFNet: A Computationally Efficient Source Filter Model Based Neural Speech Synthesis
Achuth Rao, M. V., & Ghosh, P. K. (2020) · IEEE Signal Processing Letters · doi:10.1109/lsp.2020.3005031 · Cited by 7
“Similar to [1], we use l1 loss between the ground truth log Mel-spectrum and the generated log Mel-spectrum to train the network in an end-to-end manner1.”
MethodTu, T., Chen, Y.-J., Liu, A. H., & Lee, H.-Y. (2020) · Interspeech (USB) · doi:10.21437/interspeech.2020-1824 · Cited by 12
Creating Song From Lip and Tongue Videos With a Convolutional Vocoder
Zhang, J., Roussel, P., & Denby, B. (2021) · IEEE Access · doi:10.1109/access.2021.3050843 · Cited by 8
“Griffin-Lim is similar conceptually to the algorithm in [32], and is also the synthesis technique employed in the generative Tacotron vocoder [35].”
MethodThe NUS & NWPU system for Voice Conversion Challenge 2020
Tian, X., Wang, Z., Shan, Y., Zhou, X., Du, H., Zhou, Y. X., Zhang, M., Zhou, K., Şişman, B., Xie, L., et al. (2020) · doi:10.21437/vcc_bc.2020-26 · Cited by 13
Liu, D., Wang, C., Gong, H., Ma, X., Tang, Y., & Pino, J. (2022) · Interspeech (USB) · doi:10.21437/interspeech.2022-10568 · Cited by 11
“[16, 17]), the output speech duration is a by-product of sourcetarget attention and is less straightforward to control.”
BackgroundInfluentialEnd-to-End Spoken Language TranslationPreprint
Guo, M., Haque, A., & Verma, P. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1904.10760 · Cited by 7
“Our method operates on spectrograms and combines the Listen-Attend-Spell model (Chan et al., 2016) with the TacoTron (Wang et al., 2017) architecture.”
MethodResultInfluentialVerma, P., & Salisbury, J. K. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.04076 · Cited by 8
“Various architectures from vision and natural processing have been successfully deployed to solve various problems in audio transformations [4], style transfer [5], speech recognition [6], learning latent representations [7], speech to speech translation [8], synthesis [9], perception [10].”
Backgrounds-Transformer: Segment-Transformer for Robust Neural Speech SynthesisPreprint
Wang, X., Ming, H., He, L., & Soong, F. K. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.08480 · Cited by 7
LaughNet: synthesizing laughter utterances from waveform silhouettes and a single laughter examplePreprint
Luong, H.-T., & Yamagishi, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.04946 · Cited by 7
“To push the limitation of the technology further, many studies focused on balancing between developing a seamless end-to-end (E2E) system [1] and putting back some form of control over synthetic speech [2].”
BackgroundDeep Fake Detection, Deterrence and Response: Challenges and OpportunitiesPreprint
Azmoodeh, A., & Dehghantanha, A. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2211.14667 · Cited by 7
“At the core of Tacotron, a module was responsible for extracting new representations from inputted voice sequence.”
MethodInfluentialArmanTTS single-speaker Persian datasetPreprint
Shamgholi, M. H., Saeedi, V., Peymanfard, J., Alhabib, L., & Zeinali, H. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2304.03585 · Cited by 7
“Tacatron[1], TransformerTTS[2], and Fast Speech[15] are examples of the end-to-end model.”
BackgroundAUDDT: A Unified Benchmark Toolkit for Audio and Speech Deepfake DetectorsPreprint
Zhu, Y., Guimarães, H. R., Pimentel, A., & Falk, T. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.21597 · Cited by 7
“Some representative models that follow this type of design include variants of Tacotron [27], [28] and the FastSpeech series [29], [30], to name just a few.”
Applications and Advances of Artificial Intelligence in Music Generation:A Review
Chen, Y., Huang, L., & Gou, T. (2024) · arXiv · doi:10.48550/arxiv.2409.03715 · Cited by 6
A Tunable Forced Alignment System Based on Deep Learning: Applications to Child Speech
Kadambi, P., Mahr, T., Hustad, K. C., & Berisha, V. (2025) · Journal of Speech Language and Hearing Research · doi:10.1044/2024_jslhr-24-00347 · Cited by 7
ArFake: A Robust Framework for Multi-Dialect Arabic Speech Spoofing Detection Benchmark
Elsetohy, M., Ehab, A., Mekky, A., Hassan, B., & Shehata, S. (2026) · Interspeech (USB) · doi:10.21437/interspeech.2026-2665 · Cited by 7
A Comprehensive Survey of DeepFake Generation and Detection Techniques in Audio-Visual Media
Khan, I. N., Khan, K., & Ahmad, A. (2025) · ICCK Journal of Image Analysis and Processing · doi:10.62762/jiap.2025.431672 · Cited by 12
“Advanced speech synthesis techniques, including Tacotron [154], Deep Voice [20], and WaveNet [112], have made it easier to produce highly realistic fake voices.”
DisCo_Speech: Controllable Zero-Shot Speech Generation with A Disentangled Speech Codec
Li, T., Ge, W., Wang, Z., Cui, Z., Ma, Y., Gao, Y., Deng, C., Zhang, S., & Feng, J. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.863 · Cited by 8
A Learned Conditional Prior for the VAE Acoustic Space of a TTS System
Karanasou, P., Karlapati, S., Moinet, A., Joly, A., Abbas, A., Slangen, S., Lorenzo-Trueba, J., & Drugman, T. (2021) · Interspeech (USB) · doi:10.21437/interspeech.2021-528 · Cited by 6
Hi robot, it’s not what you say, it’s how you say it
Miniota, J., Wang, S., Beskow, J., Gustafson, J., Székely, É., & Pereiral, A. (2023) · IEEE RO-MAN · doi:10.1109/ro-man57019.2023.10309427 · Cited by 6
“Recent advancements in speech synthesis using neural models have made it possible to create voices that can sometimes sound indistinguishable from human voices [1].”
BackgroundXue, J., Deng, Y., Han, Y., Gao, Y., & Li, Y. (2024) · Interspeech (USB) · doi:10.21437/interspeech.2024-1056 · Cited by 10
VQ-CTAP: Cross-Modal Fine-Grained Sequence Representation Learning for Speech Processing
Qiang, C., Wang, G., Zhao, Y., Fu, R., Wang, T., Cheng, G., Wang, T., Liu, Q., Yi, J., Wen, Z., et al. (2025) · IEEE Transactions on Audio Speech and Language Processing · doi:10.1109/taslpro.2025.3564168 · Cited by 12
“Traditional TTS methods have achieved satisfactory results [27], [28], [29], [30], [31], [32].”
Janyoi, P., & Seresangtakul, P. (2020) · Applied Sciences · doi:10.3390/app10186381 · Cited by 6
Paul, D., Mukherjee, S., Pantazis, Y., & Stylianou, Y. (2021) · Interspeech (USB) · doi:10.21437/interspeech.2021-660 · Cited by 6
“Recently, the superiority of deep neural network (DNN) based speech synthesis surpassed the conventional speech synthesis models [1, 2, 3, 4, 5, 6].”
BackgroundEmotional Speech Cloning using GANs
Selvi, S. S., Anantharamakrishnan, V., Koushik, A., & Akhil, S. K. (2021) · International Conference on Recent Trends on Electronics, Information, Communication & Technology (RTEICT) · doi:10.1109/rteict52294.2021.9573848 · Cited by 6
“Wang et al [6] developed TacoTron, a spectrogram synthesizer that converts character inputs into spectrograms.”
Background
Metadata
- Authors
- Wang, Y., Skerry-Ryan, R. J., Stanton, D., Wu, Y., Weiss, R. J., Jaitly, N., Yang, Z., Xiao, Y., Chen, Z., Bengio, S., Le, Q., Agiomyrgiannakis, Y., Clark, R., & Saurous, R. A.
- Affiliations
- Google (United States) (US)
- Corresponding author
- Wang, Yuxuan
- ORCID
- Venue
- Interspeech 2017
- Publisher
- ISCA
- Published
- 2017-08-20
- Pages
- 4006-4010
- ISSN
- 2308-457X (ISSN-L)
- OpenAlex
- W2963609956
- Cited by
- 1,746
- References
- 25
- Citations by year
- 2026: 134 · 2025: 187 · 2024: 183 · 2023: 215 · 2022: 243 · 2021: 301 · 2020: 252 · 2019: 176 · 2018: 45 · 2017: 9
- FWCI
- 60.12
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- Interspeech (USB) (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
- Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
- Health Sciences › Health Professions › Speech and Hearing › Voice and Speech Disorders
- Keywords
- Tacotron, end-to-end speech synthesis, text-to-speech, sequence-to-sequence models, naturalness
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.