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-citationSignal estimation from modified short-time Fourier transform
Griffin, D., & Lim, J. (1984) · IEEE Transactions on Acoustics, Speech, and Signal Processing · doi:10.1109/tassp.1984.1164317 · Cited by 2,095
Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks
Bengio, S., Vinyals, O., Jaitly, N., & Shazeer, N. (2015) · arXiv · doi:10.48550/arxiv.1506.03099 · Cited by 1,621 · cited 2×
“Since we do not use techniques such as scheduled sampling [20] (we found it to hurt audio quality), the dropout in the pre-net is critical for the model to generalize, as it provides a noise source to resolve the multiple modalities in the output distribution.”
MethodSelf-citationStatistical parametric speech synthesis
Zen, H., Tokuda, K., & Black, A. W. (2009) · Speech Communication · doi:10.1016/j.specom.2009.04.004 · Cited by 1,195 · cited 2×
“For example, it is common for statistical parametric TTS to have a text frontend extracting various linguistic features, a duration model, an acoustic feature prediction model and a complex signal-processing-based vocoder (Zen et al., 2009; Agiomyrgiannakis, 2015).”
MethodA note on the evaluation of generative models
Theis, L., Oord, A. V. D., & Bethge, M. (2015) · arXiv · doi:10.48550/arxiv.1511.01844 · Cited by 1,026 · cited 1×
“hard to compare models based on objective metrics, which often do not correlate well with perception [24].”
BackgroundSrivastava, R. K., Greff, K., & Schmidhuber, J. (2015) · arXiv · doi:10.48550/arxiv.1505.00387 · Cited by 912 · cited 2×
“CBHG consists of a bank of 1-D convolutional filters, followed by highway networks [15] and a bidirectional gated recurrent unit (GRU) [16] recurrent neural net (RNN).”
BackgroundVinyals, O., Kaiser, L., Koo, T., Petrov, S., Sutskever, I., & Hinton, G. (2014) · arXiv · doi:10.48550/arxiv.1412.7449 · Cited by 464 · cited 4×
“[14]), where a stateful recurrent layer produces the attention query at each decoder time step.”
BackgroundMethodInfluentialFully Character-Level Neural Machine Translation without Explicit Segmentation
Lee, J., Cho, K., & Hofmann, T. (2017) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00067 · Cited by 394 · cited 3×
“CBHG is inspired from work in machine translation [8], where the main differences from [8] include using non-causal convolutions, batch normalization, residual connections, and stride=1 max pooling.”
MethodInfluentialSampleRNN: An Unconditional End-to-End Neural Audio Generation Model
Mehri, S., Kumar, K., Gulrajani, I., Kumar, R., Jain, S., Sotelo, J., Courville, A. C., & Bengio, Y. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1612.07837 · Cited by 392 · 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.”
MethodInfluentialDeep Voice: Real-time Neural Text-to-Speech
Arık, S. Ö., Chrzanowski, M., Coates, A., Diamos, G., Gibiansky, A., Kang, Y., Li, X., Miller, J. J., Ng, A. Y., Raiman, J., et al. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1702.07825 · Cited by 381 · 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.”
MethodInfluential(2017) · International Conference on Learning Representations · Cited by 314
Zen, H., Agiomyrgiannakis, Y., Egberts, N., Henderson, F., & Szczepaniak, P. (2016) · Interspeech (USB) · doi:10.21437/interspeech.2016-522 · Cited by 121 · cited 5×
“We compare our model with a parametric (based on LSTM (Zen et al., 2016)) and a concatenative system (Gonzalvo et al., 2016), both of which are in production.”
MethodInfluentialSelf-citationRNN Approaches to Text Normalization: A ChallengePreprint
Sproat, R., & Jaitly, N. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1611.00068 · Cited by 110 · cited 2×
“We perform simple text normalization, though recent advancements in learned text normalization [26] may render this unnecessary in the future.”
MethodSelf-citationVocaine the vocoder and applications in speech synthesis
Agiomyrgiannakis, Y. (2015) · IEEE International Conference on Acoustics, Speech, and Signal Processing · doi:10.1109/icassp.2015.7178768 · Cited by 76 · cited 2×
“For example, it is common for statistical parametric TTS to have a text frontend extracting various linguistic features, a duration model, an acoustic feature prediction model and a complex signal-processing-based vocoder (Zen et al., 2009; Agiomyrgiannakis, 2015).”
MethodSelf-citationWang, W., Xu, S., & Xu, B. (2016) · Interspeech (USB) · doi:10.21437/interspeech.2016-134 · Cited by 64
Recent Advances in Google Real-Time HMM-Driven Unit Selection Synthesizer
Gonzalvo, X., Tazari, S., Chan, C.-A., Becker, M., Gutkin, A., & Silén, H. (2016) · Interspeech · doi:10.21437/interspeech.2016-264 · Cited by 62 · cited 2×
“We compare our model with a parametric (based on LSTM (Zen et al., 2016)) and a concatenative system (Gonzalvo et al., 2016), both of which are in production.”
Method
Cited by
Natural TTS Synthesis by Conditioning Wavenet on MEL Spectrogram Predictions
Shen, J., Pang, R., Weiss, R. J., Schuster, M., Jaitly, N., Yang, Z., Chen, Z., Zhang, Y., Wang, Y., Skerrv-Ryan, R., et al. (2018) · 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2018.8461368 · Cited by 2,791
Self-citationIEEE Symposium on Security and Privacy
(2023) · IT Professional · doi:10.1109/mitp.2023.3246667 · Cited by 1,167
Data Augmentation Generative Adversarial Networks
Antoniou, A., Storkey, A., & Edwards, H. S. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1711.04340 · Cited by 747
“Waveglow: a flow-based generative network for speech synthesis,”
Prenger, R., Valle, R., & Catanzaro, B. (2019) · ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2019.8683143 · Cited by 1,004
Deep Learning for Audio Signal Processing
Purwins, H., Li, B., Virtanen, T., Schluter, J., Chang, S.-Y., & Sainath, T. (2019) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2019.2908700 · Cited by 855
“via the Griffin-Lim algorithm [65] in combination with the inverse Fourier transform [139] which does not reach high synthesis”
Method
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.