tts_transformer-zh-cv7_css10

Transformer text-to-speech model from fairseq S^2 (paper/code):

• Simplified Chinese
• Single-speaker female voice
• Pre-trained on Common Voice v7, fine-tuned on CSS10

Usage

from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub<br /> from fairseq.models.text_to_speech.hub_interface import TTSHubInterface<br /> import IPython.display as ipd<br /> models, cfg, task = load_model_ensemble_and_task_from_hf_hub(<br /> "facebook/tts_transformer-zh-cv7_css10",<br /> arg_overrides={"vocoder": "hifigan", "fp16": False}<br /> )<br /> model = models[0]<br /> TTSHubInterface.update_cfg_with_data_cfg(cfg, task.data_cfg)<br /> generator = task.build_generator(model, cfg)<br /> text = "您好，这是试运行。"<br /> sample = TTSHubInterface.get_model_input(task, text)<br /> wav, rate = TTSHubInterface.get_prediction(task, model, generator, sample)<br /> ipd.Audio(wav, rate=rate)<br />

See also fairseq S^2 example.

Citation

@inproceedings{wang-etal-2021-fairseq,<br /> title = "fairseq S{\^{}}2: A Scalable and Integrable Speech Synthesis Toolkit",<br /> author = "Wang, Changhan and<br /> Hsu, Wei-Ning and<br /> Adi, Yossi and<br /> Polyak, Adam and<br /> Lee, Ann and<br /> Chen, Peng-Jen and<br /> Gu, Jiatao and<br /> Pino, Juan",<br /> booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",<br /> month = nov,<br /> year = "2021",<br /> address = "Online and Punta Cana, Dominican Republic",<br /> publisher = "Association for Computational Linguistics",<br /> url = "https://aclanthology.org/2021.emnlp-demo.17",<br /> doi = "10.18653/v1/2021.emnlp-demo.17",<br /> pages = "143--152",<br /> }<br />