ACE Studio features

Tracks

Track types

A track has no fixed type, and by default it is a blank track. It becomes what you put on it — load a singing voice and it is a vocal track, load an instrument and it is an instrument track. A project starts with a large pool of empty tracks waiting to be filled.

• A MIDI track with no sound source is not silent — it falls back to a standard piano sample library. When the user says "it sounds like a piano", the answer is almost always that nothing is loaded, not that something broke.
• Every track except a blank track can host audio effects.
• Unison Mode belongs to voice tracks and AI instrument tracks only. It lets several voices or instruments load onto one track to produce a unison. Open the spread up when you want a wider stereo image.

Audio effects

• ACE Studio ships basic effects, but prefer the third-party plugin effects the user has installed locally.
• Voice tracks and instrument tracks come with a basic effect chain already loaded. Add to it or strip it back as the work requires.
• Room Effect is a spatial effect that exists to serve Unison Mode: it models one acoustic space and places the unison speakers at different positions within it. Turn it on only when the track plays a background role.
• To take an effect out of the chain, deactivate (disable) it — do not bypass it. Bypass is for comparing a plugin in context, a listening call an agent does not make.

Vocal Synth, AI Instruments, and instrument plugins

Vocal Synth

Use Vocal Synth when the user wants to create a vocal. It is the sound source that takes MIDI and lyrics, and synthesises a sung vocal through a chosen speaker.

Recommended speakers:

Blend voice creates a new voice by mixing several speakers. When the user describes a voice they want, pick from the library first; reach for a blend only if the library does not get there and they want something bespoke.

If the user wants their own voice, or a specific voice they supply, open the voice library and point them at the Cloning tab to train it themselves. That flow happens in a browser. Consult the wiki and walk them through it.

Vocal Synth renders in the cloud, but there is a turbo mode that moves most of the model computation onto the local machine. Studio measures whether the machine can support turbo mode the first time it launches: if the local compute is enough it is on by default, and if the check fails it cannot be enabled. Turbo mode turns on pre-rendering by default, so an edit triggers a render immediately. With turbo mode and pre-rendering together, rendering is usually finished before playback even starts.

• Without pre-rendering, rendering is triggered by playback.
• With pre-rendering, rendering is triggered by the edit.

AI Instrument

When the user wants an acoustic instrument, prefer an AI Instrument — its realism is far beyond a sample library.

AI Instruments do not render in real time, so each instrument type carries a very small sample pack purely so MIDI input responds as you play. That sample is not what the model outputs.

String Section is modelled on a string orchestra. Its speakers are Violins I, Violins II, Violas, Celli and Basses. These support polyphony well, but keep the written part idiomatic for the section it is on.

AI Instruments render in the cloud, with no turbo mode and no pre-rendering, so rendering is always triggered by playback.

Every other instrument is a solo instrument. You can feed them polyphonic material, but they do not behave the way a sample library does:

• Strings tolerate a little polyphony — but as one instrument bowing several strings, not as a section.
• Wind instruments cannot physically play polyphonically, so the model is forced to fake it. Do not do this.
• When a solo instrument needs several parts, split them across tracks instead.

Instruments named Vintage perform less well than the others.

Articulations are marked on notes, and each instrument type has its own set. The default smart mode works out the articulation from the melody and applies it in the synthesised result. Two articulations never appear under smart mode and must be set explicitly:

• Pizzicato
• Mute

Rendering — Vocal Synth and AI Instruments alike

Whatever triggers a render, it only starts on a track that can currently be heard. A muted track makes no sound, and neither does any track while another track is soloed — and a silent track never renders. If nothing is rendering, check mute and solo before anything else.

During playback, a region that has not finished rendering blocks playback.

Instrument plugins

ACE Studio supports third-party VST3 and AU instrument plugins.

When the user wants a particular sound:

• If it is an acoustic instrument, look through the AI Instrument list first.
• Then look through the user's own plugin list for something suitable.

Once a plugin is loaded, it is fully drivable — parameters, whole state and presets, and its own interface through Studio's built-in computer use. See ace-studio-audio-plugins for how to drive it, and for per-plugin recipes.

Generative Kits

Every generative kit takes a prompt and produces audio. Generation takes over a minute. There are three:

• Inspire Me — text to song. Use this when the user wants to generate a whole song outright.

• Music Enhancer — AI cover. Select a region of the timeline as the input context. Once the tool has uploaded the audio it returns the lyrics and style it recognised; edit either (or neither) to produce the cover. Three common uses:

  • Style replacement. Keep the lyrics, rewrite the style description, turn one finished piece of music into another.
  • Finishing an idea. Record a hum or write a melodic motif, then write the lyrics and style out in full to grow the motif into a complete demo.
  • Refinement. Keep the lyrics and style as they are and regenerate. The content usually shifts slightly, but it turns a rough demo into a polished one.
• Add A Layer — generation as splice. Select the region to generate into. The tool uploads the surrounding timeline content as reference audio and generates the specified sample from the prompt. The result fits what is already on the timeline — a bass line, a drum part, a lush string passage.

How to prompt Add A Layer

Choose custom mode. Then write two short prompts, one per field:

• Instrument — the instrument and its timbre.
• Styles — the musical content: style, emotional colour, layering and dynamics, and the groove and physical feel of the playing.

Keep both short. Do not describe rhythm or chords — the surrounding audio is uploaded as reference, so the model already has the harmony and the groove; restating them adds nothing. And do not pile up negative prompts: this model is built for synthesising a single timbre, so by default it does not need to be told what to avoid.

Utilities

Stem Splitter

Four ways to split:

• Vocal, instrumental.
• Vocal, drums, bass, piano, guitar, others. Nothing is distinguished within a category — several guitar tracks may all land in the guitar stem — and some instruments end up in Others.
• All detected stems. Separation is not guaranteed to be complete; instruments of the same category often share a stem, and some land in FX or Others.
• Customized. Describe what to separate in a prompt. This suits non-musical audio better; on musical content the granularity is coarse — ask for hi-hat and you will likely get the whole drum kit. The operation produces two tracks: the separated target, and everything left over.

Voice Changer

Converts a monophonic vocal on the timeline into a vocal or instrument audio in a chosen timbre.

A set of preset models is included to cover a wide range of creative needs, and the surface lists every model a conversion can use. When the user needs a timbre outside it, guide them to train their own Voice Changer model.

Training input is usually a vocal, but in practice it also accepts instrument audio with a clear timbre — which must also be monophonic.

Vocal to MIDI

Recognises a sung vocal in audio as a MIDI clip with lyrics. Supports monophonic vocals; it cannot correctly resolve stacked, multi-part vocals.

The V2 model can pull the lead vocal's lyrics and melody straight out of a complete song, and holds up well against background noise.

Expect some notes to come back at the wrong pitch, and fix them before the clip is used. A single note is not enough to judge — a drifting note and a deliberate accidental look the same in isolation. Read the melody as a whole, work out what key it is actually in, and correct the notes that fall outside it.

Doubles

Vocal Synth tracks and AI Instrument tracks support doubles.

The result is two duplicated tracks whose notes carry a degree of random offset from the originals. The two new tracks are panned left and right.

Marker track

A marker track can be created at the top of the timeline to give the user reference annotations against time — sections, lyrics, and so on.

Tempo track

When the music changes tempo, draw tempo automation on the tempo track. Note that editing the tempo track overwrites globally.

When the user wants the project tempo synced to imported audio, use the Apply Tempo interface to apply the audio's tempo to the project. Note that this operation shifts the target audio later so that the audio's first downbeat lands on the nearest bar line.

Importing files

Studio imports the common formats onto the timeline — MIDI, audio, MusicXML and video among them. The surface's docs carry the full extension list.