Piano recording to editable MIDI
A 15-second authorized piano recording. The result contains 191 note events. Download audio and MIDI to check pitch and timing. Notes are unchanged from the result. 191 is a file count, not accuracy. No finished score is produced.
https://melumimi.com/en/midi-result-check
Source / attribution: MeluMimi authorized piano example. Original audio and uncorrected MIDI: linked case page. Visualization: actual MIDI note events.

Hear the music, vocals and backing
One 30-second music excerpt, with actual vocal and instrumental outputs. Compare and download the input and output WAV files. One example is not a promise for every song. Check leakage, residual sound and tone loss. No artist endorsement.
https://melumimi.com/en/stem-separation-real-test
Source / attribution: Source: Stumble Bee — Lay down the law, excerpt 00:30–01:00. CC0 1.0. https://freemusicarchive.org/music/stumble-bee/single/lay-down-the-law/ | https://creativecommons.org/publicdomain/zero/1.0/ . Waveforms derived from the existing public input and vocal/instrumental outputs. No artist endorsement.

Speech cleanup: listen for detail
The same 30-second reading with added synthetic noise. Compare actual standard and gentle outputs with the reference and input. A controlled added-noise example, not a real street interview or benchmark. Check residual noise and voice damage.
https://melumimi.com/en/audio-denoising-real-test
Source / attribution: Source: LibriSpeech © 2014 Vassil Panayotov, prepared with Daniel Povey. Reader: Nolan Fout, Ten Years Later, 11 - Night, recordings 6930-75918-0004–0007. CC BY 4.0. https://www.openslr.org/12/ | https://creativecommons.org/licenses/by/4.0/ . Existing example concatenates and excerpts the reading, resamples it, adds synthetic noise and hum, and denoises the input. These visualizations derive from that existing input and actual outputs; audio is not bundled. No reader or dataset-author endorsement.