Input and Output Files

Input and Output Files

Input files

The file chooser supports the following common file extensions:

Family	Extensions shown by the picker
MPEG audio	.mp3
Waveform audio	.wav
Lossless audio	.flac
Ogg	.ogg
MPEG-4/AAC	.m4a, .aac
Windows Media Audio	.wma
Audio Interchange File Format	.aiff, .aif

The picker also offers All Files. This list is a convenience filter, not a guarantee that every codec variant or damaged file with one of these extensions can be decoded.

What really determines support

A2M first asks SoundFile to decode the selected file. If that fails, it tries audioread. Actual support therefore depends on:

• the audio data inside the file, not only its extension;
• the codecs supported by SoundFile/libsndfile;
• the fallback decoder and backend available in the packaged application;
• whether the file contains valid sample-rate and channel metadata;
• whether the decoder can return complete audio samples.

A renamed file does not become a supported format. Conversely, All Files can be useful when a valid audio file has an unusual extension.

Decoder troubleshooting

If a file fails before transcription starts:

1. Confirm that it plays normally in another local application.
2. Convert a copy to PCM WAV or FLAC using a trusted audio editor.
3. Avoid changing only the filename extension.
4. Check that the copy contains actual audio and is not zero-length or truncated.
5. Retry in CPU mode to separate decoding problems from GPU-runtime problems.

Channel and sample-rate handling

Decoded audio is represented with an explicit channel dimension. A2M accepts both mono and multichannel input, replaces non-finite sample values with finite bounded values, and resamples the audio to the sampling rate required by the active model. The current model operates at 44,100 Hz.

During feature extraction, A2M processes all available channels and combines their spectral information. The pipeline should therefore not be described as selecting a single channel or unconditionally downmixing the input to mono.

Output files

A2M writes Standard MIDI data with:

• one instrument named A2M Piano;
• the Acoustic Grand Piano MIDI program;
• note pitches clamped to MIDI range 0 through 127;
• note velocities clamped from 1 through 127;
• 960 ticks-per-quarter-note resolution in the PrettyMIDI representation;
• optional sustain and soft-pedal controller events.

The current piano model produces notes in the acoustic piano range, MIDI pitches 21 through 108. Its modeled pedal symbols are sustain, CC64, and soft pedal, CC67. It does not currently produce sostenuto CC66.

Output location and naming

The default folder is Downloads\A2M, but Settings can point to another folder. A2M creates the selected output folder when possible, sanitizes the source stem for Windows, and reserves a unique filename before transcription writes the MIDI.

If conversion fails or is canceled after that reservation, A2M removes the reserved or partial result.