| ## Example to download fastspeech2 from fairseq | |
| The following should work with fairseq's most up-to-date version in a google colab: | |
| ```python | |
| from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub | |
| import IPython.display as ipd | |
| import torch | |
| model_ensemble, cfg, task = load_model_ensemble_and_task_from_hf_hub( | |
| "facebook/fastspeech2-en-ljspeech", arg_overrides={"vocoder": "griffin_lim", "fp16": False} | |
| ) | |
| 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 = "This is a cool demo for speech synthesis, don't you think so?" | |
| 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"] | |
| ipd.Audio(waveform, rate=task.sr) | |
| ``` | |
| See: https://colab.research.google.com/drive/1gvq4Y1urrg9QrQ9031sZIP93LKspIh_X?usp=sharing |