Labyricorn/voicebox
1
0
Fork 0
mirror of https://github.com/jamiepine/voicebox.git synced 2026-09-15 04:40:40 -07:00
Code Issues Packages Projects Releases Wiki Activity

Revert "fix: cast librosa float64 audio to float32 for Chatterbox voice encoder"

Browse Source 
This reverts commit 1d32170c2e.
This commit is contained in:
James Pine
2026-03-13 04:28:00 -07:00
parent 1d32170c2e
commit 5ccf79a8f7
2 changed files with 0 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/backends/chatterbox_backend.py
-24
View File
@@ -178,30 +178,6 @@ class ChatterboxTTSBackend:
progress_manager.mark_complete(model_name)
task_manager.complete_download(model_name)
# Monkey-patch prepare_conditionals to cast librosa's float64
# audio to float32 before it reaches the voice encoder. The
# upstream VoiceEncoder's melspectrogram only casts to float32
# when hp.normalized_mels is True (it defaults to False), so
# float64 mels hit float32 model weights → dtype mismatch.
_orig_prepare = self.model.prepare_conditionals
def _f32_prepare(wav_fpath, **kwargs):
import librosa as _librosa
_orig_librosa_load = _librosa.load
def _load_f32(*a, **kw):
wav, sr = _orig_librosa_load(*a, **kw)
return wav.astype(np.float32), sr
_librosa.load = _load_f32
try:
return _orig_prepare(wav_fpath, **kwargs)
finally:
_librosa.load = _orig_librosa_load
self.model.prepare_conditionals = _f32_prepare
logger.info("Chatterbox Multilingual TTS loaded successfully")
except ImportError as e:
backend/backends/chatterbox_turbo_backend.py
-24
View File
@@ -178,30 +178,6 @@ class ChatterboxTurboTTSBackend:
progress_manager.mark_complete(model_name)
task_manager.complete_download(model_name)
# Monkey-patch prepare_conditionals to cast librosa's float64
# audio to float32 before it reaches the voice encoder. The
# upstream VoiceEncoder's melspectrogram only casts to float32
# when hp.normalized_mels is True (it defaults to False), so
# float64 mels hit float32 model weights → dtype mismatch.
_orig_prepare = self.model.prepare_conditionals
def _f32_prepare(wav_fpath, **kwargs):
import librosa as _librosa
_orig_librosa_load = _librosa.load
def _load_f32(*a, **kw):
wav, sr = _orig_librosa_load(*a, **kw)
return wav.astype(np.float32), sr
_librosa.load = _load_f32
try:
return _orig_prepare(wav_fpath, **kwargs)
finally:
_librosa.load = _orig_librosa_load
self.model.prepare_conditionals = _f32_prepare
logger.info("Chatterbox Turbo TTS loaded successfully")
except ImportError as e: