-
-
Notifications
You must be signed in to change notification settings - Fork 0
Troubleshooting
Justagwas edited this page Jul 15, 2026
·
1 revision
Start with the exact message A2M shows. Provider and package validation errors are intentionally specific.
| Symptom | Likely cause | Verification | Next action |
|---|---|---|---|
| File cannot be opened | Neither SoundFile nor audioread could decode it | Try a short known-good WAV | Re-export to PCM WAV or FLAC |
| Model download repeats | No valid bundle passes manifest/component validation | Check model folder and the displayed download error | Retry on a stable connection; report persistent hash/validation failures |
| Conversion button remains unavailable | File, model, worker, or runtime preflight is not ready | Check model status and active task | Finish/stop the current setup task and select a valid file |
| Output folder error | Folder cannot be created or written | Choose a user-writable directory | Change Output Location |
| GPU pack downloads but is rejected | Provider, ABI, metadata, layout, or helper validation failed | Keep the exact reason text | See the provider-specific page |
| A2M requests restart | Native runtime backend changed | Confirm the selected provider | Restart before conversion |
| GPU failure changes mode to CPU | A recognized GPU error triggered fallback state | Check the status label | Start the file again in CPU mode |
| Conversion seems stuck at a section | Long inference, resource pressure, or backend problem | Watch whether progress eventually advances and check system memory | Stop, lower resource settings, retry CPU |
| MIDI has missing/extra notes | Ambiguous acoustic evidence | Compare difficult passages in a piano roll | See Accuracy, Limitations and Best Results |
| Note starts are good but durations are not | Releases and pedal resonance are harder to infer | Compare note endings separately | Edit releases and pedals; evaluate their timing separately |
| Stop is not immediate | Cancellation is checked at safe boundaries | Wait for the current decode/inference operation | Do not terminate unless it remains unresponsive |
When reporting a problem, include:
- A2M version;
- CPU/GPU mode and selected provider;
- the exact displayed reason;
- whether the provider status says CPU, CUDA, or DirectML;
- Windows version, GPU model, and driver;
- input format and approximate duration;
- whether the same file works in CPU mode;
- whether the issue happens after an A2M restart.
Do not attach private audio publicly unless you have permission. A short generated or freely licensed reproducer is preferable.