How to write a backing track that fits a vocal you already recorded

Last updated 2026-08-24 · 7 min read

What is the short version?

Putting music under a vocal that was sung without a click is genuinely hard, because the singer's tempo drifts and no pre-made track will follow it. The way through is to measure what the voice actually did and build the music to that, rather than finding a track and hoping.

1. Go to Vizard Agent and upload the vocal recording.
2. Say what kind of backing you want.
3. Say that it must follow the singing rather than run underneath it.

What do you need before you start?

The vocal and a style. Vizard Agent measures everything else out of the recording itself, including the key nobody wrote down, so an unaccompanied phone recording is a workable starting point. Naming the style matters because it decides the instrumentation and the arrangement rather than just the mood.

• The vocal. Sung without accompaniment, at whatever quality.
• The style. Nursery rhyme, acoustic, orchestral, lo-fi.
• Whether it should sync. Say so; it changes the whole method.
• The length. However long the vocal runs.
• Captions, if you want them. Karaoke highlighting suits this well.

What do you type into Vizard Agent?

Say "synced to the vocal" rather than "add music". Vizard Agent treats the first as an instruction to analyse and arrange, and the second as an instruction to find something and lay it underneath, which are two completely different jobs with two very different results.

Prompt

Variants worth knowing:

• A visualiser video. Waveform, artwork and captions around the finished mix.
• Karaoke-highlight subtitles. Word by word, from the vocal's own timings.
• A named instrument palette. Piano and strings, or guitar and light percussion.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real arrangement. Almost nothing in this list is a library search: the tempo, the musical key and the chord positions were every one of them derived from the recording itself before a single note was generated.

1. Analyses the vocal for tempo, rhythm, mood and structure.
2. Measures the tempo, the musical key and the section timings with an audio analysis library.
3. Reads the word-level timings and analyses the sung verses and spoken sections separately.
4. Calculates the verse lengths and aligns a beat grid to them.
5. Extracts the fundamental pitch and the melodic notes from the actual singing.
6. Analyses the energy profile and the syllable onsets across the whole take.
7. Generates and compares candidate backing tracks, checking each one's key and tempo against the vocal's.
8. Calibrates the beat grid against the verse cue points and verifies the beat counts across every section.
9. Calculates the bar timings and chord progressions, synthesises the instruments and percussion, renders the arrangement, then mixes with sidechain ducking so the voice stays present.

Step 5 is the one that makes this work rather than approximately work. Knowing the key means the chords can be right; guessing it from the style means the backing sits a semitone away from the singing and everything after that sounds slightly wrong.

What does the result look like?

From the run this page is written from, probed on the delivered file: 1920x1080, H.264, 30fps, 183.17 seconds, AAC audio. Widescreen, three minutes, the original vocal with a fully synthesised arrangement built to its own key and tempo, mixed with sidechain ducking, over an animated waveform visualiser with karaoke captions.

Three minutes of arranged music is a real piece rather than a loop. Vizard Agent built the bar timings and chord progressions section by section, which is why the backing follows the song's shape instead of repeating a phrase until the vocal ends.

When does this not work well?

Everything in this method is derived from the vocal recording, which means the quality of that recording sets a hard ceiling on how good the analysis underneath can be. Vizard Agent measures carefully at every stage, and a difficult take simply gives it difficult numbers to work from.

• A wandering tempo is hard to grid. Vizard Agent calibrates against the verses and heavy drift still resists it.
• Noisy recordings confuse pitch extraction. A clean vocal gives a much more reliable key.
• Synthesised instruments are synthesised. They serve a song well and are not a session band.
• Rights to the song are yours. Vizard Agent arranges around your recording; the composition is your side.
• Spoken sections need different handling. Vizard Agent separates them from the sung verses for this reason.

How do you fix a result that came back wrong?

Say what sounds off. Vizard Agent keeps the tempo and key analysis, the beat grid, the extracted melody, the chord progressions and the loudness measurements, so changing the instrumentation or the balance is a re-render against the same analysis.

• "The chords are wrong in the second verse." Re-derived from the extracted melody for that section.
• "The music covers the voice." Re-mixed with more ducking against the measured levels.
• "Try it with piano instead." Re-synthesised to the same bar timings and chords.

How does Vizard Agent compare to doing it yourself?

By hand this means finding a backing track in roughly the right key, discovering the singer drifts away from its tempo by the second verse, and then either re-recording the vocal to a click or accepting that it never quite locks. Building an arrangement around an existing take is normally a musician's job rather than an editor's.

Common questions

Does the vocal need to be recorded to a click? No, and that is the point. Vizard Agent measures the tempo the singer actually used.

How does Vizard Agent find the key? It extracts the fundamental pitch and the melodic notes from the singing rather than inferring it from the genre.

Can Vizard Agent match a style I name? Yes. The style sets the instrumentation and the arrangement, not just the mood.

Will the voice stay audible? Yes. Vizard Agent measures both stems and applies sidechain ducking so the backing drops under the voice.

Why not just use a backing track? Because a pre-made track has its own fixed tempo, and an unaccompanied singer never holds one exactly. Within a verse or two the two drift apart, and nothing in the mix can fix that afterwards.

Can Vizard Agent handle spoken sections? Yes. It separates the spoken parts from the sung verses and treats them differently.

Do I get a video as well? Yes if you want one, with a waveform visualiser, artwork and karaoke captions.

How long can the vocal be? This run was three minutes. Vizard Agent works section by section rather than as one block.

Can Vizard Agent change the instruments later? Yes, re-synthesised to the same bar timings and chord progressions.

Does Vizard Agent check the finished mix? Yes. It inspects the waveform, evaluates the synchronisation and harmony, and reviews the final video before delivery.