Own & train

Turn your own playing into an AI instrument model

Pick at least 10 minutes of solo recordings of an instrument you play and train a model of its sound. Then hum a line in Voice to Instrument and hear it played back as your instrument.

Train Instrument Model builds an AI instrument model from at least 10 minutes of recordings of an instrument you play. It is for players and producers who want their own instrument's sound, such as a cello or a sitar, playable with their voice. When training finishes, hum or sing a line in Voice to Instrument and hear it come back as your instrument.

See it in action

How it works

  1. Name your instrument

    Open Train Instrument Model under Build & Train in AI Tools and type an Instrument name, such as My Nylon Guitar.

  2. Add your recordings

    Pick at least 10 minutes of solo, dry recordings from your Library, each at least 60 seconds, covering low to high notes played both soft and loud.

  3. Set quality and visibility

    Slide Training quality from 10 epochs (Fast) to 350 (Studio); 180 is the default. Keep the model Private or make it Public, then turn on the consent switch.

  4. Train, then play it

    Press Start Training and follow each epoch in the chat. When it finishes, your model is under My DNA and in Voice to Instrument, under Yours.

Why use Train Instrument Model

Your instrument, your sound

The model learns from recordings of the instrument you play, so it captures that instrument's timbre instead of one of the built-in presets.

Play it with your voice

In Voice to Instrument, hum or sing a melody and hear it come back in the sound of the instrument you trained.

Choose your training depth

Slide from 10 to 350 epochs. More epochs give a higher-fidelity model but take longer to train and use more tokens.

Watch it train

The chat shows the upload and each epoch, and Voice to Instrument lists the run as Training until it is ready to pick.

Great for

  • Play your own cello with your voice
  • Capture a rare instrument you play
  • Sketch parts in your instrument's sound
  • Hum melodies into your own guitar tone
  • Share your instrument in the Public tab
  • Keep one instrument sound across tracks

Frequently asked questions

What is Train Instrument Model?

Train Instrument Model is the Soundverse tool that turns at least 10 minutes of recordings of an instrument you play into an AI instrument model. Soundverse adds the finished model to Voice to Instrument, where you hum or sing a melody and hear it played back in your instrument's sound.

What recordings do I need to train an instrument model?

Train Instrument Model needs at least 10 minutes of solo, dry, close-miked recordings of one instrument, with each clip at least 60 seconds long. Cover the whole range from low to high notes at different strengths, and include sustained notes and articulations so the model captures the instrument's true timbre. Train Instrument Model takes MP3, WAV, FLAC and M4A files from your Library.

How do I play my instrument model after training?

Open Voice to Instrument, add a hummed or sung melody, and choose your trained model under the Yours tab. Voice to Instrument maps the pitch and rhythm of your line onto the instrument you trained, so a hummed phrase comes back as your cello or guitar. Models you trained with Train Instrument Model also appear under My DNA.

How long does it take to train an instrument model?

Train Instrument Model runs the same background training as Train Voice Model, so a model is ready hours after you press Start Training, not minutes. Higher Training quality settings take longer, from 10 epochs at the Fast end to 350 at Studio. The chat shows each epoch, and Voice to Instrument lists the run as Training until it is ready.

Is Train Instrument Model free?

Train Instrument Model is available on the Creator plan and above, and each training run uses tokens from your plan. The cost depends on the epochs you choose, and the panel shows it before you start. Soundverse is free to sign up, and paid plans add a commercial license for what you make; the Pricing page lists what each plan covers.

Can I train a model on recordings I didn't make?

Only if you are licensed to use them. Train Instrument Model asks you to confirm “I own these recordings or am licensed to use them”, and Start Training stays disabled until you switch it on. Recordings of your own playing are the simplest and safest source. You need the rights to anything you upload.

How is an instrument model different from a voice model?

Train Instrument Model and Train Voice Model run the same training with the same epoch range and visibility settings; the main difference is the target. An instrument model learns from solo recordings of an instrument and plays your hummed melodies in Voice to Instrument, while a voice model learns from a voice and re-sings vocals in Voice Swap.

What are the limits of an instrument model?

Train Instrument Model learns only from the recordings you give it, so cover the instrument's full range at different strengths; background music, noise or other instruments in the takes lower quality. It doesn't compose or arrange by itself: it gives Voice to Instrument a new sound for your melodies. Built-in presets such as sitar, bansuri and flute need no training at all.

Ready to train your instrument?

Add at least 10 minutes of solo recordings, pick the training quality and start training.