Practice a song section with your own audio

1. Open an audio file

Choose “Choose an audio file” in open practice. Use a file you have permission to use, up to 100 MB: MP3, M4A, WAV or another format your browser supports. The file and its name stay on your device. A Spotify or YouTube link is not an audio file this player can open.

2. Mark the part you want to repeat

Use the player to find your section. Choose “Set start here” and “Set end here”, or enter the start and end in seconds. Then turn on “Repeat a section”. The end must be after the start and within the track. You can also try a slower playback speed.

3. Put on headphones and start the microphone

Choose “Start microphone” and allow browser microphone access. Use the audio player’s play button to start your track. Keep this page visible and sing comfortably. Speaker playback can be mistaken for your voice; the app does not remove the original singer.

4. Read the line as an estimate

The note name is the nearest musical note to the sound the microphone captured, not the note you were supposed to sing. A gap means no estimate was returned. The detector can miss real singing or show a wrong pitch. It cannot tell you whether you sang this song correctly.

5. Use a known target when you want a comparison

Switch to “Guided pitch exercise” for an original three-note reference and estimated high/low feedback. This is separate from your chosen song. You can also switch to “Listen and repeat only” to try the phrase without feedback. Learning benefits remain unvalidated.

No local audio file?

You can try playing music separately through headphones and then starting the microphone here. Device and app behavior varies: another app may interrupt capture. Keep this page visible. Our section-repeat and speed controls apply only to a file opened in our player. We do not download or analyze Spotify or YouTube recordings.

A gap in the line is not a grade for your voice

The open mode estimates the pitch reaching the microphone. It has no target melody for your song. A gap means the tool did not return an estimate; it can miss real singing too. Keep your voice comfortable instead of trying to force a line to appear.

What our recording check found

On 25 September 2026, we checked the shipped detector against all 40 solo-singing recordings in vocadito v3, using independent pitch labels. We did not change their volume or remove difficult clips.

75.3% of labelled singing frames received an estimate within half a semitone. Missing estimates count as misses.

22.4% of labelled singing frames received no estimate.

6.0% of frames labelled unvoiced received a pitch estimate anyway.

17 / 40 clips missed our per-clip reliability gate. The overall result also missed our 85% gate.

This is a small offline test, not a test of your phone, noisy karaoke, improved singing or customer demand. Overlapping frames are not independent people. These recordings are not our customer sessions.

Method and full results (English)

Summary data (JSON) · All 40 clips (CSV)

Original dataset · Bittner and colleagues, 2021 · CC BY 4.0

Tell us what helped or stopped you