Efficient Synthesis of Violin Sounds Using a BiLSTM Network Based Source Filter Model

Dai YR, Yang HC, Su AWY. Efficient Synthesis of Violin Sounds Using a BiLSTM Network Based Source Filter Model. In: AES Convention 150. Audio Engineering Society; 2021. Paper 10476. Available from: https://aes.org/publications/elibrary-page/?id=21069

@inproceedings{Dai2021_21069,
  author = {Dai, Yi-Ren and Yang, Hung-Chih and Su, Alvin W.Y.},
  title = {{Efficient Synthesis of Violin Sounds Using a BiLSTM Network Based Source Filter Model}},
  booktitle = {AES Convention 150},
  note = {Paper 10476},
  year = {2021},
  month = may,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=21069}
}

TY  - CPAPER
TI  - Efficient Synthesis of Violin Sounds Using a BiLSTM Network Based Source Filter Model
AU  - Dai, Yi-Ren
AU  - Yang, Hung-Chih
AU  - Su, Alvin W.Y.
T2  - AES Convention 150
M1  - Paper 10476
PY  - 2021
DA  - 2021/05/06
UR  - https://aes.org/publications/elibrary-page/?id=21069
PB  - Audio Engineering Society
LA  - en
AB  - The dynamic changes in playing skills generated from bow-string interaction make synthesizing bowed string instrument sounds a difficult task. Recently, a source filter model incorporating the LSTM predictor and the granular wavetables gives encouraging results. However, the prediction error is still large and the model hasn’t caught the nuance caused by the constantly changing characteristics of a playing violin. In this paper, the granular wavetable is represented of DCT coefficients and a new training strategy is proposed to reduce the predictor error. In addition, we analyze the difference between the original violin tone and the corresponding synthesis tone. A random pitch perturbation and a DCT coefficient shaping method are proposed to imitate the changing characteristics since results sound regular.
ER  -