James Pine bfd7b815a5 fix: patch S3Tokenizer.log_mel_spectrogram for float64→float32 cast ...

The actual dtype mismatch was in S3Tokenizer.log_mel_spectrogram, not
VoiceEncoder.forward. librosa.load returns float64 numpy, which
torch.from_numpy preserves as double. The STFT output (double) then
hits _mel_filters (float32) in a matmul at s3tokenizer.py:163.

Now patching both entry points after model load:
1. S3Tokenizer.log_mel_spectrogram — cast audio to float32 before STFT
2. VoiceEncoder.forward — cast mels to float32 before LSTM

Remove debug traceback logging (no longer needed).

2026-03-13 05:04:29 -07:00

..

__init__.py

feat: add Chatterbox Turbo engine and per-engine language lists

2026-03-13 02:35:10 -07:00

chatterbox_backend.py

fix: patch S3Tokenizer.log_mel_spectrogram for float64→float32 cast

2026-03-13 05:04:29 -07:00

chatterbox_turbo_backend.py

fix: patch S3Tokenizer.log_mel_spectrogram for float64→float32 cast

2026-03-13 05:04:29 -07:00

luxtts_backend.py

fix: download progress tracking for all engines and inline progress UI

2026-03-13 02:09:32 -07:00

mlx_backend.py

fix: pass language parameter to Qwen TTS models and sync form with profile language

2026-03-13 04:04:04 -07:00

pytorch_backend.py

fix: pass language parameter to Qwen TTS models and sync form with profile language

2026-03-13 04:04:04 -07:00