Update README.md

README.md

@@ -9,40 +9,59 @@ language: en
 datasets:
 - ljspeech
 ---
-## Example to download TTS Transformer from fairseq
+# tts_transformer-en-ljspeech
 
-The following should work with fairseq's most up-to-date version in a google colab:
+[Transformer](https://arxiv.org/abs/1809.08895) text-to-speech model from [fairseq S^2](https://github.com/pytorch/fairseq/tree/main/examples/speech_synthesis):
+- English
+- Single-speaker female voice
+- Trained on [LJSpeech](https://keithito.com/LJ-Speech-Dataset/)
+
+## Usage
 
 ```python
 from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub
+from fairseq.models.text_to_speech.hub_interface import TTSHubInterface
 import IPython.display as ipd
-import torch
 
-model_ensemble, cfg, task = load_model_ensemble_and_task_from_hf_hub(
-    "facebook/tts_transformer-en-ljspeech", arg_overrides={"vocoder": "griffin_lim", "fp16": False}
+
+models, cfg, task = load_model_ensemble_and_task_from_hf_hub(
+    "facebook/tts_transformer-en-ljspeech",
+    arg_overrides={"vocoder": "hifigan", "fp16": False}
 )
+model = models[0]
+TTSHubInterface.update_cfg_with_data_cfg(cfg, task.data_cfg)
+generator = task.build_generator(model, cfg)
 
-def tokenize(text):
-  import g2p_en
-  tokenized = g2p_en.G2p()(text)
-  tokenized = [{",": "sp", ";": "sp"}.get(p, p) for p in tokenized]
-  return " ".join(p for p in tokenized if p.isalnum())
-  
 text = "Hello, this is a test run."
 
-tokenized = tokenize(text)
-sample = {
-    "net_input": {
-        "src_tokens": task.src_dict.encode_line(tokenized).view(1, -1),
-        "src_lengths": torch.Tensor([len(tokenized.split())]).long(),
-        "prev_output_tokens": None
-        },
-    "target_lengths": None,
-    "speaker": None,
-}
-generator = task.build_generator(model_ensemble, cfg)
-generation = generator.generate(model_ensemble[0], sample)
-waveform = generation[0]["waveform"]
+sample = TTSHubInterface.get_model_input(task, text)
+wav, rate = TTSHubInterface.get_prediction(task, model, generator, sample)
+
+ipd.Audio(wav, rate=rate)
+```
+
+See also [fairseq S^2 example](https://github.com/pytorch/fairseq/blob/main/examples/speech_synthesis/docs/ljspeech_example.md).
 
-ipd.Audio(waveform, rate=task.sr)
+## Citation
+
+```bibtex
+@inproceedings{wang-etal-2021-fairseq,
+    title = "fairseq S{\^{}}2: A Scalable and Integrable Speech Synthesis Toolkit",
+    author = "Wang, Changhan  and
+      Hsu, Wei-Ning  and
+      Adi, Yossi  and
+      Polyak, Adam  and
+      Lee, Ann  and
+      Chen, Peng-Jen  and
+      Gu, Jiatao  and
+      Pino, Juan",
+    booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
+    month = nov,
+    year = "2021",
+    address = "Online and Punta Cana, Dominican Republic",
+    publisher = "Association for Computational Linguistics",
+    url = "https://aclanthology.org/2021.emnlp-demo.17",
+    doi = "10.18653/v1/2021.emnlp-demo.17",
+    pages = "143--152",
+}
 ```