Can you lose an unsaved MP3 to MIDI result?

The 2003 TallStick TS-AudioToMIDI quick-start help described a sequence that separated the source audio, a result held inside the program, and a MIDI file written to disk. Those were not interchangeable states.

First, the user opened a WAVE file. The historical instructions said that choosing the audio enabled the program's playback and conversion controls. At this stage there was an input file, but no claim that a MIDI file had already been created or saved.

Next, the user started recognition. When that process finished, the program placed the created MIDI result in its built-in sequencer. The user could play the result there. Being able to hear it therefore proved that recognition had produced something the program could hold and audition; it did not by itself prove that a separate MIDI file existed on the device.

The distinction became unusually clear in the vendor's description of the unregistered program. TallStick said that this edition allowed playback of the generated result but disabled the save command. In other words, the same result could be available for preview while unavailable as a saved file. That restriction belonged to this historical product and licence state. It is not evidence that present-day converters use the same restriction.

The archived help also described what happened when a user changed recognition settings and tried again. Running recognition another time replaced the MIDI already held by the program. The vendor therefore told users to save a result they wanted to keep before starting another pass. That is a lifecycle warning, not an accuracy claim: it concerns which result remained available, not whether either pass contained the right notes.

This yields a precise three-state model. The WAVE file was the source. The sequencer-held MIDI was the latest in-program result. A saved MIDI file was the persistent copy outside that replaceable result slot. Losing the middle state did not imply that the source audio vanished, and saving the third state did not alter the source. Each object had a different role.

The archive does not say how many passes a user should try, how often results were lost, or whether one setting produced better transcription. Those numbers are unavailable, so none is supplied here. The supported lesson is narrower: preview access was not proof of persistence, and repeat recognition could replace the unsaved result.

This site's converter builds the MIDI in your browser and gives the result view a download link. That link points to a temporary object URL created from the generated MIDI data. When you choose Convert another, the page releases that object URL as it resets the result view.

The behavior follows the browser mechanism documented in the W3C File API: an object URL can refer to a Blob, and software can revoke that URL. Revoking the page's link is different from deleting a file that your browser has already downloaded. The practical boundary is the completed download, not the fact that the result was visible on screen.

1. Convert a short excerpt. A small test makes it easier to recognise which pass you are checking.
2. Use Download MIDI. Do this while the result view and its download link are still present.
3. Confirm the file exists. Open the browser's download list or the destination folder, then import the file into your DAW or notation editor.
4. Name the pass clearly. Add a setting or purpose to the filename in your file manager or DAW so the next result cannot be confused with it.
5. Only then convert again. Return to the tool, choose Convert another, and create the next result.

This procedure tests persistence without assuming how every converter stores its workspace. If the downloaded file opens after the page resets, you have an independent copy to compare. If you only heard a preview and never exported it, treat that pass as temporary until the tool proves otherwise.

For an A/B comparison, download the first result and import it into one track. Create the second result, download it, and place it on another track aligned to the same audio excerpt. Mute one MIDI track at a time. This lets you compare note timing and content without relying on a preview that may have been replaced.

Do not assume that a later pass is automatically better. Preserve both files, listen against the same source, and keep the one that needs less correction for your purpose.