mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-09-03 06:04:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

minor updates related to the tokenizer change

Browse Source
This commit is contained in:
yaozengwei 2024-02-28 20:37:42 +08:00
parent 595d4a3c47
commit ae83d8070d
3 changed files with 9 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
egs/ljspeech/TTS/vits/export-onnx.py
View File

@ -218,8 +218,7 @@ def main():
params.update(vars(args))
tokenizer = Tokenizer(params.tokens)
params.blank_id = tokenizer.blank_id
params.oov_id = tokenizer.oov_id
params.blank_id = tokenizer.pad_id
params.vocab_size = tokenizer.vocab_size
logging.info(params)

9
egs/ljspeech/TTS/vits/infer.py
View File

@ -130,14 +130,16 @@ def infer_dataset(
batch_size = len(batch["tokens"])
tokens = batch["tokens"]
tokens = tokenizer.tokens_to_token_ids(tokens)
tokens = tokenizer.tokens_to_token_ids(
tokens, intersperse_blank=True, add_sos=True, add_eos=True
)
tokens = k2.RaggedTensor(tokens)
row_splits = tokens.shape.row_splits(1)
tokens_lens = row_splits[1:] - row_splits[:-1]
tokens = tokens.to(device)
tokens_lens = tokens_lens.to(device)
# tensor of shape (B, T)
tokens = tokens.pad(mode="constant", padding_value=tokenizer.blank_id)
tokens = tokens.pad(mode="constant", padding_value=tokenizer.pad_id)
audio = batch["audio"]
audio_lens = batch["audio_lens"].tolist()
@ -201,8 +203,7 @@ def main():
device = torch.device("cuda", 0)
tokenizer = Tokenizer(params.tokens)
params.blank_id = tokenizer.blank_id
params.oov_id = tokenizer.oov_id
params.blank_id = tokenizer.pad_id
params.vocab_size = tokenizer.vocab_size
logging.info(f"Device: {device}")

4
egs/ljspeech/TTS/vits/test_onnx.py
View File

@ -108,7 +108,9 @@ def main():
model = OnnxModel(args.model_filename)
text = "I went there to see the land, the people and how their system works, end quote."
tokens = tokenizer.texts_to_token_ids([text])
tokens = tokenizer.texts_to_token_ids(
[text], intersperse_blank=True, add_sos=True, add_eos=True
)
tokens = torch.tensor(tokens) # (1, T)
tokens_lens = torch.tensor([tokens.shape[1]], dtype=torch.int64) # (1, T)
audio = model(tokens, tokens_lens) # (1, T')