speech_recognition/FunASR
1
0
Fork 0
mirror of https://github.com/modelscope/FunASR synced 2025-09-15 14:48:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

auto frontend

Browse Source
This commit is contained in:
游雁 2024-06-07 03:05:01 +08:00
parent b2be308de0
commit 4eedd7ad45
1 changed files with 12 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
funasr/datasets/openai_datasets/datasets.py
View File

@ -201,16 +201,16 @@ class OpenAIDataset(torch.utils.data.Dataset):
data_list, batch_first=True, padding_value=pad_value
)
# if self.batch_type != "example":
# for i in range(10):
# outputs = self._filter_badcase(outputs, i=i)
if self.batch_type != "example":
for i in range(10):
outputs = self._filter_badcase(outputs, i=i)
return outputs
def _filter_badcase(self, outputs, i=0):
b, t, _ = outputs["speech"].shape
b, t = outputs["input_ids"].shape
if b * t > self.batch_size * 1.25:
if b * t > self.batch_size * 2:
beg = torch.randint(0, 2, ()).item()
if b < 2:
beg = 0
@ -219,12 +219,12 @@ class OpenAIDataset(torch.utils.data.Dataset):
)
for key, data_list in outputs.items():
outputs[key] = outputs[key][beg : beg + b : 2]
speech_lengths_max = outputs["speech_lengths"].max().item()
outputs["speech"] = outputs["speech"][:, :speech_lengths_max, :]
text_lengths_max = outputs["text_lengths"].max().item()
outputs["text"] = outputs["text"][:, :text_lengths_max]
target_mask_lengths_max = outputs["target_mask_lengths"].max().item()
outputs["target_mask"] = outputs["target_mask"][:, :target_mask_lengths_max]
#
# speech_lengths_max = outputs["speech_lengths"].max().item()
# outputs["speech"] = outputs["speech"][:, :speech_lengths_max, :]
# text_lengths_max = outputs["text_lengths"].max().item()
# outputs["text"] = outputs["text"][:, :text_lengths_max]
# target_mask_lengths_max = outputs["target_mask_lengths"].max().item()
# outputs["target_mask"] = outputs["target_mask"][:, :target_mask_lengths_max]
return outputs