Karaoke Trainer

How to use Karaoke Trainer

Karaoke Trainer shows the original singer's melody as a line and draws your pitch over it while you sing, so you can see where you went flat or sharp. This guide walks through each part of the screen.

Start in three steps

  1. Load a song. Tap “Choose a song file” and pick a song you have (MP3, WAV, M4A, FLAC…). Wait until the bar is full.
  2. Put on headphones and press “Sing”. The song starts and your pitch is drawn over the grey melody line.
  3. Press “Stop singing”. See how much was on pitch, and tap a part that went off to sing it again.

No song at hand? Press “Try the demo song”: a short song made for this app, which needs no download.

Loading a song

Choose a file, or drop it on the box. Any audio format your browser can play works (MP3, WAV, M4A (AAC), FLAC, Ogg…), up to 10 minutes. The app then:

  1. separates the vocals from the accompaniment with an AI model (Spleeter), and
  2. finds the melody of the separated vocals: the pitch every 10 milliseconds.

The bar shows how far it is and about how long is left. On a recent computer a 4-minute song is usually ready in a minute or two; a phone takes longer. Keep the screen on until it finishes.

The first time, the browser downloads the AI model (about 75 MB) and the engine that runs it (about 14 MB, or 27 MB when the GPU is used): about 89 MB in all, only once. On a phone or tablet the app asks before it starts, so you can wait for Wi-Fi.

Every song you load is kept on this device (the last 8). Under “Recent songs”, tap one and it opens at once, with no separation and even offline. Choosing the same file again also opens the saved copy.

The app only opens files on your device. It does not take songs from YouTube or streaming services. Use songs you bought or recorded, for your own practice.

Reading the melody line

The practice screen while singing: a piano keyboard on the left with note names, the grey melody line of the demo song, and the singer's line drawn over it: green where it is on pitch, yellow and red where it is lower than the melody. A black vertical line marks now, and the note being sung shows at the top right.
The demo song while singing. Tap the image to enlarge it.

Singing

  1. Put on headphones. With a speaker, the microphone hears the song too, and the original singer's voice gets mixed into your line. Wired headphones are best; Bluetooth headphones add a delay (see Settings).
  2. Press “Sing”. The first time, the browser asks to use the microphone: allow it. The song starts from where it is.
  3. Sing along. Your line appears as you sing.
  4. Press “Stop singing” (or pause, or let the song end) to see how you sang.

Octaves: many people sing a song an octave lower or higher than the original singer (a man singing a song by a woman, for example). That counts as on pitch. The app finds your octave in the first seconds you sing and draws your line next to the melody; the menu next to “Sing” shows what it found (“Octave: auto (lower)”), and you can choose the octave yourself.

Looking back: how you sang

How you sang: the share of the time on pitch, the share of the melody sung, the average in cents (a minus sign for lower), and a list of the parts that went off, each with its time and how many cents low or high. Below, buttons to compare the original vocal and my voice, and to save my singing with the song (MP3) or my voice only (WAV), with a note that the MP3 is for personal use only.
After singing the demo song (the demo song has no “Keep this recording” button).

The MP3 is for your personal use only. It contains the song, and the rights to the song belong to its rights holders. Use it to listen back to your own practice, and do not upload it to video or social media sites, send it to others, or distribute it. The app asks you to confirm this each time, and writes it into the file's tag (comment).

The numbers are a guide for practice, not a score. The line and the list tell you more: where you go flat at the end of a long note, which jump you overshoot, which phrase starts late.

Vocal volume and key

Finding your key (optional)

If a song feels too high or too low, measure your vocal range once (“Measure my range” in the song panel, or ⚙ → “My vocal range”). You do not need it to practice.

  1. Sing the lowest note you can sing comfortably on “ah” and hold it for 4 seconds.
  2. Then the highest note in your normal voice (not falsetto).
  3. Optionally, the highest note in falsetto.

The range is shown on a keyboard and saved on this device. From then on, each song shows a suggestion next to the key, such as “Suggested for you: −3”. Tap it to use that key. The app compares the melody's range (its low and high notes, leaving out the lowest and highest 5% of the time so a stray note does not count) with your range, and picks the change that fits best; sometimes the best fit is the original key sung an octave lower or higher.

Settings: how strict, delay, microphone

⚙ → Settings.

How strictGreen (cents)Yellow (cents)
Easy±75±150
Normal (default)±50±100
Strict±25±50

Delay correction: every device takes a moment to play the sound and to hear the microphone (Bluetooth headphones much more). The app subtracts this delay. If your line is drawn later than the melody even when you sing in time, press “Measure the delay” and say “ta” with each of the 8 beeps, or move the slider by hand.

Microphone: noise suppression, echo cancellation, and automatic gain are off by default, because they can cut or change parts of the voice and make the pitch harder to follow. Turn on echo cancellation only if you must use a speaker.

On a phone or tablet

Everything works on a phone, in one column: after a song is loaded, the screen scrolls to the melody line. Loading a new song takes longer than on a computer; keep the screen on until it is ready. On iPhone and iPad, the sound plays even with the silent switch on. The app can be added to the home screen and then works offline.

What is kept on your device

The AI model, the songs you loaded (the separated vocals and accompaniment and the melody line), recordings you chose to keep, your vocal range if you measured it, and your settings. Nothing is uploaded. Delete them with ⚙ → “Delete saved songs”, ⚙ → “Delete the AI model”, and “Delete” in “My vocal range”. See the privacy policy.