-
-
Notifications
You must be signed in to change notification settings - Fork 0
Settings Reference
Settings affect execution resources, MIDI export, or the interface. They do not select different transcription models.
| Setting | Choices / range | Actual effect |
|---|---|---|
| Device mode | CPU, GPU | Chooses CPU execution or the selected validated GPU provider |
| GPU provider | CUDA, DirectML | Selects the managed ONNX Runtime provider requested in GPU mode |
| CPU usage | Low, Balanced, High, Max | Uses about 25%, 50%, 75%, or 100% of logical processors as the worker/thread target |
| GPU memory usage | Low, Balanced, High, Max | Maps to a hardware-bounded section batch size; it is not a precise VRAM reservation |
| Pedal events | Off, On | Excludes or writes decoded pedal events in the MIDI |
| Dynamics | Expressive, Uniform | Keeps model-predicted velocity or replaces every note velocity |
| Uniform velocity | 1 through 127, default 96 | Value used only in Uniform mode |
| UI size | 75 through 200% | Scales the application interface |
| Output location | Folder | Directory used for newly written MIDI files |
| Automatic updates | On, Off | Enables the startup update check |
The CPU worker targets are calculated with ceiling rounding and at least one worker. “50%” therefore means a target derived from logical processors, not a promise that Task Manager will show exactly 50%.
GPU mode still uses CPU work for decoding, feature construction, post-processing, and MIDI writing. Its CPU usage choice and GPU batch choice control different parts of the pipeline.
If a GPU scorer batch fails, A2M reduces the batch and retries. The lower working batch is cached for that active model/provider session.
New configuration defaults are CPU, DirectML as the preferred GPU provider, pedals on, expressive dynamics, Balanced performance, 100% UI scale, automatic update checks on, and Downloads\A2M as the output folder. First-run hardware detection may initialize performance and GPU-batch defaults based on logical processors and detected memory.
Settings are normalized before saving to %LOCALAPPDATA%\A2M\a2m_config.json. Current schema version: 7.
Source: config.py, config_service.py, resource_service.py, settings_panel.py, and app_controller.py.