Skip to content

fix: prevent sample loss in chunked_decode #259

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
zhuxiaoxuhit wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix: prevent sample loss in chunked_decode #259

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
7 changes: 6 additions & 1 deletion qwen_tts/core/tokenizer_12hz/modeling_qwen3_tts_tokenizer_v2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -891,7 +891,12 @@ def chunked_decode(self, codes, chunk_size=300, left_context_size=25):
context_size = left_context_size if start_index - left_context_size > 0 else start_index
codes_chunk = codes[..., start_index - context_size : end_index]
wav_chunk = self(codes_chunk)
wavs.append(wav_chunk[..., context_size * self.total_upsample :])
# The causal decoder may produce fewer samples than context_size * total_upsample
# due to implicit delays in causal convolutions. Taking the last sample_count samples
# from wav_chunk is more robust than slicing from a fixed offset, which can
# over-remove samples when the decoder output is shorter than expected.
sample_count = min((end_index - start_index) * self.total_upsample, wav_chunk.shape[-1])
wavs.append(wav_chunk[..., -sample_count:])
start_index = end_index
return torch.cat(wavs, dim=-1)

Expand Down