malibrated 4ead99754a Minor improvements to account for large-VRAM Apple Silicon machines, and refine the inferencing code to ignore spurious, non-audio responses from the Orpheus model, use the system prompt if it exists in the directory structure, and for long texts, 1) use nltk for segmenting the text according to user-specified max character lengths, and 2) use the previous segment (For segment index n>1) as context in the system prompt, to hopefully improve audio generation consistency.		2025-04-21 22:30:33 -07:00
..
__init__.py	Minor improvements to account for large-VRAM Apple Silicon machines, and refine the inferencing code to ignore spurious, non-audio responses from the Orpheus model, use the system prompt if it exists in the directory structure, and for long texts, 1) use nltk for segmenting the text according to user-specified max character lengths, and 2) use the previous segment (For segment index n>1) as context in the system prompt, to hopefully improve audio generation consistency.	2025-04-21 22:30:33 -07:00
inference.mps	Minor improvements to account for large-VRAM Apple Silicon machines, and refine the inferencing code to ignore spurious, non-audio responses from the Orpheus model, use the system prompt if it exists in the directory structure, and for long texts, 1) use nltk for segmenting the text according to user-specified max character lengths, and 2) use the previous segment (For segment index n>1) as context in the system prompt, to hopefully improve audio generation consistency.	2025-04-21 22:30:33 -07:00
inference.py	Minor improvements to account for large-VRAM Apple Silicon machines, and refine the inferencing code to ignore spurious, non-audio responses from the Orpheus model, use the system prompt if it exists in the directory structure, and for long texts, 1) use nltk for segmenting the text according to user-specified max character lengths, and 2) use the previous segment (For segment index n>1) as context in the system prompt, to hopefully improve audio generation consistency.	2025-04-21 22:30:33 -07:00
speechpipe.py	Minor improvements to account for large-VRAM Apple Silicon machines, and refine the inferencing code to ignore spurious, non-audio responses from the Orpheus model, use the system prompt if it exists in the directory structure, and for long texts, 1) use nltk for segmenting the text according to user-specified max character lengths, and 2) use the previous segment (For segment index n>1) as context in the system prompt, to hopefully improve audio generation consistency.	2025-04-21 22:30:33 -07:00