Skip to content

fix(asr): align acoustic and semantic feature lengths to prevent tensor mismatch #309

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
JasonOA888 wants to merge 1 commit into
base: main
Choose a base branch
from
+17 −0
Open

fix(asr): align acoustic and semantic feature lengths to prevent tensor mismatch #309

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 17 additions & 0 deletions vibevoice/modular/modeling_vibevoice_asr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -261,6 +261,14 @@ def encode_speech(
else:
semantic_tokens = self.model.semantic_tokenizer.encode(speech_tensors.unsqueeze(1)).mean
semantic_features = self.model.semantic_connector(semantic_tokens)

# Ensure acoustic and semantic features have matching temporal dimensions.
# Different tokenizer architectures can produce slightly different
# frame counts, causing tensor size mismatches on concatenation.
min_len = min(acoustic_features.shape[1], semantic_features.shape[1])
if acoustic_features.shape[1] != semantic_features.shape[1]:
acoustic_features = acoustic_features[:, :min_len]
semantic_features = semantic_features[:, :min_len]
else:
# Long audio: streaming processing
# print(f"Using streaming processing for long audio: {total_samples/sample_rate:.1f}s "
Expand Down Expand Up @@ -329,6 +337,15 @@ def _iter_segments(total_length: int, segment_length: int):
# Concatenate all semantic means
semantic_tokens = torch.cat(semantic_mean_segments, dim=1).contiguous()
semantic_features = self.model.semantic_connector(semantic_tokens)

# Ensure acoustic and semantic features have matching temporal dimensions.
# Different segment boundaries can cause slight frame-count mismatches
# between the two tokenizers (e.g. tensor a (228) vs tensor b (223)).
min_len = min(acoustic_features.shape[1], semantic_features.shape[1])
if acoustic_features.shape[1] != semantic_features.shape[1]:
acoustic_features = acoustic_features[:, :min_len]
semantic_features = semantic_features[:, :min_len]
semantic_features = self.model.semantic_connector(semantic_tokens)

# Combine acoustic and semantic features
if speech_masks is not None:
Expand Down