nodes/diodiogod-TTS-Audio-Suite
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix Step Audio EditX edit node sample rate mismatch causing pitch shift

Browse Source 
CRITICAL: Audio editor node was saving input audio with engine's sample rate (24000 Hz)
regardless of actual input sample rate (44100 Hz, 48000 Hz, etc.), causing pitch shift.

Changes:
- Track current sample rate through iterations
- Resample to 24000 Hz before saving if needed
- Update sample rate after each iteration (engine returns 24000 Hz)
- Handle cached iterations at correct sample rate

Fixes pitch preservation when editing external audio files.
This commit is contained in:
diodiogod
2025-12-05 08:28:46 -03:00
parent a55b3bca20
commit d7359e55b3
1 changed files with 21 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nodes/step_audio_editx_special/step_audio_editx_audio_editor_node.py
+21 -1
View File
@@ -354,12 +354,14 @@ class StepAudioEditXAudioEditorNode:
# Find highest cached iteration to resume from
start_iteration = 0
current_tensor = audio_tensor
current_sample_rate = sample_rate # Track current sample rate
for i in range(n_edit_iterations, 0, -1):
if i in cached_iterations:
print(f"💾 CACHE: Resuming from cached iteration {i}/{n_edit_iterations}")
current_tensor = cached_iterations[i]
start_iteration = i
current_sample_rate = step_audio_engine.get_sample_rate() # Cached tensors are at engine sample rate
break
# Save initial/resumed audio to temp file for engine
@@ -392,7 +394,22 @@ class StepAudioEditXAudioEditorNode:
save_tensor = save_tensor.unsqueeze(0)
elif save_tensor.dim() == 3:
save_tensor = save_tensor.squeeze(0)
torchaudio.save(temp_audio_path, save_tensor.cpu(), step_audio_engine.get_sample_rate())
# CRITICAL: Resample to engine's sample rate (24000 Hz) if needed
# Input audio might be 44100 Hz, 48000 Hz, etc.
# Saving with wrong sample rate causes pitch shift
target_sample_rate = step_audio_engine.get_sample_rate() # 24000 Hz
if current_sample_rate != target_sample_rate:
print(f" Resampling audio from {current_sample_rate}Hz to {target_sample_rate}Hz")
# Use torchaudio for resampling
resampler = torchaudio.transforms.Resample(
orig_freq=current_sample_rate,
new_freq=target_sample_rate
)
resampled_tensor = resampler(save_tensor.cpu())
torchaudio.save(temp_audio_path, resampled_tensor.cpu(), target_sample_rate)
else:
torchaudio.save(temp_audio_path, save_tensor.cpu(), current_sample_rate)
# Perform single edit with progress tracking
current_tensor = step_audio_engine.edit_single(
@@ -407,6 +424,9 @@ class StepAudioEditXAudioEditorNode:
do_sample=do_sample
)
# Update sample rate - engine always returns 24000 Hz audio
current_sample_rate = step_audio_engine.get_sample_rate()
# Cache this iteration
self._cache_iteration_result(cache_key, iteration_num, current_tensor)