Published November 27, 2023 | Version 0.0.1
Model Open

THCHS-30 Chinese TTS with explicit duration markers

  • 1. ROR icon Chemnitz University of Technology

Description

This upload contains a TTS model which was trained on the THCHS-30 dataset using these transcriptions, which contain explicit phoneme duration markers (see below). The model is trained using tacotron-cli (v0.0.4).

The model achieves the following values on the validation set:

  • mean mel-cepstral distance: 24.58
  • mean penalty: 0.1072

Files:

  • 103500.pt
    • checkpoint after 500 epochs with a batch size of 64
  • 1-setup-env.sh
    • script to install all required tools
  • 2-create-dataset.sh
    • script to create the base dataset using public resources
  • 3-create-train-val-set.sh
    • script to create the training set and validation set
  • 4-start-training.sh
    • script to start training using Tacotron
  • 5-convert-chinese-to-ipa.sh
    • script to prepare Chinese texts for synthesis by transcribing them to IPA
  • 6-synthesize.sh
    • script to synthesize IPA transcribed text
  • example-north-wind.zip
    • contains an example passage which was synthesized using the model (speaker: D7)

The model is able to synthesize the following symbols:

  • vowels: a ɛ e ə ɚ ɤ i o u ʊ y
  • diphthongs: ai̯ au̯ ei̯ ou̯
  • consonants: f j k kʰ l m n p pʰ ɹ̩ ɻ ɻ̩ s t ts tsʰ tɕ tɕʰ tʰ w x ŋ ɕ ɥ ʂ ʈʂ ʈʂʰ
  • breaks:
    • SIL0 -> no break
    • SIL1 -> short break
    • SIL2 -> break
    • SIL3 -> long break
  • special characters: 。 ?

Vowels and diphthongs contain one of these tones:

  • ˥ -> first tone, e.g., e˥
  • ˧˥ -> second tone, e.g., e˧˥
  • ˧˩˧ -> third tone, e.g., e˧˩˧
  • ˥˩ -> fourth tone, e.g., e˥˩
  • nothing -> no tone, e.g., e

Vowels, diphthongs and consonants contain one of these duration markers:

  • ˘ -> very short, e.g., ou̯˘
  • nothing -> normal, e.g., ou̯
  • ˑ -> half long, e.g., ou̯ˑ
  • ː -> long, e.g., ou̯ː

Tones and duration markers can be combined, e.g., ə˧˥ː

Available speakers:

  • male: A9, A33, A35, B21, B34, A8, B8, C8, D8
  • female: A11, A12, A13, A14, A19, A2, A22, A23, A32, A34, A36, A4, A5, A6, A7, B11, B12, B15, B2, B22, B31, B32, B33, B4, B6, B7, C12, C13, C14, C17, C18, C19, C2, C20, C21, C22, C23, C31, C32, C4, C6, C7, D11, D12, D13, D21, D31, D32, D4, D6, D7

Example sentence:

有一次, 北风 和 太阳 正在 争论 谁 比较 有本事。

j|ou̯˧˩˧|i˥|tsʰ|ɹ̩˥˩|SIL2|p|ei̯˧˩˧|f|ə˥|ŋ|SIL0|x|ɤ˧˥|SIL0|tʰ|ai̯˥˩|j|a˧˥˘|ŋ˘|SIL0|ʈʂ|ə˥˩|ŋ|ts|ai̯˥˩|SIL0|ʈʂ|ə˥|ŋ|l|w|ə˥˩ː|nˑ|SIL0|ʂ|w˘|ei̯˧˥|SIL0|p|i˧˩˧|tɕ˘|j|au̯˥˩˘|SIL0|j|ou̯˧˩˧|p|ə˧˩˧|n|ʂ|ɻ̩˥˩|。

Notes (English)

The authors gratefully acknowledge the GWK support for funding this project by providing computing time through the Center for Information Services and HPC (ZIH) at TU Dresden.

The authors are grateful to the Center for Information Services and High Performance Computing [Zentrum fur Informationsdienste und Hochleistungsrechnen (ZIH)] at TU Dresden for providing its facilities for high throughput calculations.

Files

example-north-wind.zip

Files (344.7 MB)

Name Size Download all
md5:a3be5e5858d44ac00d543bb7673c8f52
732 Bytes Download
md5:6c4523fc68b0dcc177db189ec63d9dc8
343.6 MB Download
md5:3784afec07415a697afac0baeb3952f2
3.0 kB Download
md5:df3a48296494e451e801974bf7e209a8
15.6 kB Download
md5:83913682cdda88e3245225111ecb8ffa
1.9 kB Download
md5:d270f948c6ded9bc198332a73ebaf5da
2.8 kB Download
md5:c39bb47023a56e67cc9c92556b5fbed8
2.8 kB Download
md5:75c5e26cd77df92aa93ca96f77f95c4c
1.1 MB Preview Download

Additional details

Funding

SFB 1410 416228727
Deutsche Forschungsgemeinschaft

References

  • Wang, D., Wu, D., & Zhu, X. (2001). TCMSD: A New Chinese Continuous Speech Database. International Conference on Chinese Computing (ICCC'01), 2001.
  • Wang, D., Zhang, X., & Zhang, Z. (2015). THCHS-30: A Free Chinese Speech Corpus (arXiv:1512.01882). arXiv. http://arxiv.org/abs/1512.01882
  • 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. (2017). Tacotron: Towards End-to-End Speech Synthesis. Interspeech 2017, 4006–4010. https://doi.org/10.21437/Interspeech.2017-1452
  • Shen, J., Pang, R., Weiss, R. J., Schuster, M., Jaitly, N., Yang, Z., Chen, Z., Zhang, Y., Wang, Y., Skerrv-Ryan, R., Saurous, R. A., Agiomvrgiannakis, Y., & Wu, Y. (2018). Natural TTS Synthesis by Conditioning Wavenet on MEL Spectrogram Predictions. 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 4779–4783. https://doi.org/10.1109/ICASSP.2018.8461368
  • Prenger, R., Valle, R., & Catanzaro, B. (2019). WaveGlow: A Flow-based Generative Network for Speech Synthesis. ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 3617–3621. https://doi.org/10.1109/ICASSP.2019.8683143
  • Taubert, S. (2023). THCHS-30 - Aligned IPA transcriptions (0.0.1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.7528596
  • Taubert, S. (2023). tacotron-cli (0.0.4). Zenodo. https://doi.org/10.5281/zenodo.7543638
  • Taubert, S. (2022). waveglow-cli (0.0.1). Zenodo. https://doi.org/10.5281/zenodo.7044345