Mimicore user guide

1. Download Mimicore-setup.exe from the download page (you agree to the terms there) and run it. It installs for your Windows user only and needs no admin rights.
2. If Windows shows "Windows protected your PC", click More info › Run anyway. Mimicore is new and not yet code-signed; every update is still signed and verified by the app.
3. The app shows no terms screen: you accept the terms once, on the download page, before the installer downloads. You must be 18 or older and clone only your own voice or voices you have permission to use.
4. Choose a task in guided setup (Live, voiceover or Dub) and download only its prerequisites; the guide marks the task done when you have a real result. You can skip it and use the full menu. Instant voice cloning also powers Speak; trained voices use RVC. Model downloads are explicit and no background performance test runs. Optional Dubbing and Voice Training add-ons live in Settings.
5. Sign in with Google, Discord or email. It is free: new accounts get 30 days of Studio free (India) or an Early Bird Gift with full Studio access (Global), with no card needed. Afterward you keep the Free plan (all 28 effects, the soundboard and 60 minutes of live AI voice every day with your own voice, plus short Speak, Dub and conversion clips) or choose Studio on your account page.
6. A tour of every page (Live, Speak, Dub, Voices, Soundboard, Support, Settings) shows where everything is. After each update that changes the interface, a short What's new tour shows only what changed. Press Esc to skip a tour. The 28 voice effects work right away, even before a model finishes downloading.

The virtual microphone lets Discord, Zoom, OBS and games hear your changed voice. You set it up once.

1. Open Live › Output › Set up virtual mic. Mimicore downloads the free VB-CABLE driver from vb-audio.com, checks its signature and installs it. Click Yes when Windows asks.
2. If Windows asks for a restart, restart and open Mimicore again. The cable is selected automatically.
3. In the other app, choose CABLE Output (VB-Audio Virtual Cable) as the microphone:
   • Discord: User Settings › Voice & Video › Input Device. Turn off Discord's own Noise Suppression (Krisp) and Echo Cancellation, because they fight the changed voice.
   • OBS: add an Audio Input Capture source and select CABLE Output.
   • Zoom / Teams / Meet: Settings › Audio › Microphone.
   • Games: set the in-game voice input device. If the game has no choice, set CABLE Output as the Windows default recording device.
4. Turn on Hear myself to listen to your new voice. Use headphones so the microphone doesn't pick up the speakers.

1. On Live, pick a voice or effect from the list and press the big Start button. The list shows your trained voices and the free effects; instant voices (the gallery voices too) and mixes also run live once you add Instant live voices (Settings > Add-ons; it needs a DirectX 12 graphics card). They play about 1 second behind you, so trained voices stay the choice for the lowest delay. Without the add-on they work in Speak, Dub and file conversion only. + New voice opens the voice gallery.
2. Speak. The status line shows estimated local delay and any glitches; the engine panel shows processing time.
3. Adjust latency and quality before starting: on a voice, use the slider just below the voice name. Left reduces buffering with more frequent processing and less context; right favours smoothness and quality and adds delay you will hear. There are 15 positions, from about 80 ms to 1.2 s of buffering in 80 ms steps; keep it toward the left for calls and games and use the right-hand end only for streams or recordings where delay does not matter. Stop Live before adjusting; your choice stays fixed. Effects have no slider. Actual delay and smoothness depend on the voice, audio devices and PC.
   • Faster, slightly less accurate (a switch under the slider, off by default): matches your voice against a smaller part of its training data so each block finishes sooner on a slow or busy graphics card. It never turns on by itself; turn it off to get exactly the normal behaviour. It applies the next time you press Start.
4. If Live struggles: the status line and the Live engine panel compare processing time with the block length. Stop, move the slider to a different position, try Faster, close graphics-heavy apps (games, video editors) and use wired audio; on a laptop, plug in the charger. Low GPU use does not prove that audio deadlines are being met. If a notice says your graphics card is too busy, it shows at most once every 15 minutes and goes away by itself: it only suggests what to try and never changes the slider. Use a free effect if no AI voice is smooth on your PC. On the longer positions Mimicore keeps a larger output buffer and a small background job that stops the graphics card from slowing down between blocks; it uses a little extra graphics memory while Live runs (start Mimicore with the environment variable MIMICORE_NO_KEEPALIVE=1 if you prefer the card to rest).
5. The Live engine panel under the stage shows what the engine is doing right now: processing time, CPU, GPU, video memory, glitches and the delay.

Controls on the Live page:

• Mute silences your microphone without stopping the voice, so unmuting is instant.
• Original-voice test (bypass) passes your real voice through, to check levels.
• Add an effect layers any effect on top of your voice: radio, cave, robot and more.
• Noise suppression removes background noise (on by default).
• Automatic level makes quiet microphones louder (on by default).
• Noise gate (Off / Light / Medium / Strong) silences everything between your words: keyboard, fan, breathing.
• Voice tuning (voices, not effects) has Pitch (in semitones; +12 is one octave up) and Voice strength (how strongly the sound is pulled toward the trained voice). Both apply instantly.
• 🎯 Calibrate to my voice (voices, not effects): read one sentence and Mimicore measures your pitch and your room. It then sets the pitch and noise gate for the closest match. The settings are saved for that voice and load whenever you pick it.

Open Voices & effects.

Voice types, as they are named in the app:

• Instant: cloned from a 10–20 second recording (also the gallery voices and mixes). Works in Speak, Dub and file conversion; runs in Live with the Instant live voices add-on, about 1 second behind you.
• Trained (speech) and Trained (singing): made from longer recordings (RVC). The closest likeness and the lowest delay in Live; a trained voice imported without a known purpose is just Trained.
• Text to speech: the built-in default voice in Speak and Dub. It does not clone anyone.
• Effects: free voice characters for Live and Speak.

Instant voice (10–20 seconds)

1. Click + Instant voice and give the voice a name.
2. Record about 15–20 seconds of natural, continuous speech in a quiet room, or import a clean recording. Background noise is removed automatically.
3. Click Create voice. It's ready in under a minute and works in Speak, Dub and file conversion. Instant voices work in Speak, Dub and file conversion; for Live add the Instant live voices add-on (about 1 second behind you) or train a voice (lowest delay).

Voice gallery

Click Get the voice gallery (27 MB) once, then press ▶ to hear the voices and Add to my voices to use one. There are 19 original voices, and each is a blend of two speakers, so none copies a real person.

Mix voices

Mix voices… blends two of your instant voices into a brand-new voice. Move the slider between A and B and press Create mix. It's great for characters and for a voice that is yours alone.

Train a voice (closest likeness)

Training makes the most faithful copy of a voice from 10–30 minutes of clean recordings of one person.

• On this PC needs an NVIDIA graphics card (GTX 16 / RTX 20 series or newer, 4 GB+). Install the Voice Training add-on when asked, add your recordings, say what the voice is for (speech is included; the singing model is a small optional download), choose a quality and click Start training. One training on this PC and one in the cloud can run at the same time, and Train in the background lets you keep using Live, Speak and Dub:
  • Quick test: about 10–20 minutes.
  • Good (recommended): about 25–50 minutes.
  • Best: about 1–2 hours.
• Cloud (free) works on any PC. It trains on a free Google Colab GPU in your own Google account, with the same trainer and the same quality:
  1. Click Connect Google Drive and allow access. Mimicore can see only the files it creates.
  2. Add your recordings, choose a quality and click Upload & open Colab.
  3. In the Colab tab, press ▶, click Run anyway, and allow Google Drive access.
  4. Keep the Colab tab open. Mimicore shows the progress and installs the voice automatically when it's done. The recordings are then moved to your Drive trash, unless you chose to keep them.

You can also start cloud training from the website.

Export and import voices

Open a voice's … menu and choose Export voice… to save it as one .mimivoice file. A trained voice can include its reference recording (on by default for your own voices, off for other people's; the switch is in the dialog). On the receiving PC press Import voice…, choose the file and confirm you have the right to use the voice. Progress is shown (checking, unpacking, verifying, installing) and you can cancel; a damaged or changed file is refused and nothing is installed. If the name exists you can keep both or replace the old voice. There are no share links: send the file itself, only to people allowed to use the voice. Older voice folders import through Older voice folder… in the same dialog. You can also double-click a .mimivoice file in Windows, choose Open with Mimicore, or drop it on the Mimicore window: the same import dialog opens on that file and nothing is imported until you confirm. Files on a network location are refused; copy them to this PC first.

Other actions

• ▶ Preview plays a sample.
• Convert a file… turns a recording into that voice.
• Rename and Delete manage the voice.

28 classic effects (Robot, Demon, Chipmunk, Radio, Cave, Cathedral, 8-bit and more) run instantly on the CPU. They're free with no time limit.

• Press ▶ to hear an effect, then Use to pick it for Live.
• Apply to a file processes a recording.
• To layer an effect on top of a cloned or trained voice, use Live › Add an effect.

1. Type or paste text, or load a .txt file. Your script is saved on this PC as you type, so leaving the page or closing Mimicore never loses it.
2. Pick a voice and a language (23 languages, including Hindi, Spanish, Japanese, Arabic and Chinese).
3. Adjust Emotion, Variation and Speed (half to double speed; the pitch stays the same) if you like.
4. Choose where it goes:
   • Speak now plays it on your speakers and/or the virtual microphone, so people in your call hear it. A copy is saved.
   • Save to file renders a WAV.

Trained voices speak through the default voice, converted with the pitch matched automatically. Files are marked as AI-generated in their metadata.

Projects

Each script is a project: it has a title, your text and the voice settings you chose, saved on this PC. Rename, archive and restore projects from the project list. If a save fails, the page says so and offers Retry before you render or switch.

Audiobook Studio

Switch Speak to Audiobook Studio for long texts. You can import a PDF, a Word file or scanned pages (text is read from images on your PC), split the text into chapters and render them in your voice. AI script polish and the random narration paragraph are optional and only work when you are signed in: they send the script (or your instructions) to our server and an AI provider so a language model can write the result, so do not use them on anything private; without them Mimicore applies a built-in local cleanup, and you can undo a polish. Rendering is chapter by chapter with a journal: if the app closes or the PC restarts in the middle, reopen the project and the finished chapters are checked and kept, and the render continues from the next one.

Match an accent

For a trained voice you can add a 10 to 20 second reference recording of the person speaking naturally and switch Accent on (15 regional options appear in the training dialog as Accent / region). Speak, Dub and Convert then carry part of that accent. It is not an exact copy, and live voice keeps your own accent.

Turn a video's speech into another language, spoken in your voice. It runs fully offline.

1. Install the free Dubbing add-on (Settings › Models & add-ons, about 3.2 GB) once.
2. Open Dub a video, choose a file (MP4, MOV, MKV, M4A, MP3 or WAV), the spoken language (or Detect automatically), the target language and a voice.
3. Set Original audio: keep some of the original under the dub for music and background sounds, or remove it for a clean voice track.
4. Click Dub it. Mimicore transcribes, translates sentence by sentence, speaks each line in your voice and fits each line to its original timing.
5. You get a WAV audio track and an SRT subtitle file. Put the audio on the video in any editor (CapCut, Premiere, DaVinci Resolve, Clipchamp) and add the subtitles if you like.

Dub only content you have the rights to.

• Air horn, applause, drum roll, sad trombone and more are built in.
• Add sounds to bring in your own clips: MP3, WAV, OGG, FLAC or M4A up to 5 minutes, as many as you like.
• Click a tile to play it into the virtual microphone. Turn on Also play on my speakers to hear it yourself.
• Give each sound a hotkey, set volumes in Edit, and drag tiles to reorder them.
• Sounds play even when the live voice is off.

Settings › Hotkeys. Click a box, then press a key or combination. Hotkeys work while a game or another app has focus.

Use combinations (Ctrl + Alt + …) or keys you don't use elsewhere (F7–F12, numpad): a bound key is reserved for Mimicore while it runs. A short tone confirms each hotkey; you can turn it off.

Settings has seven tabs: Account, Hotkeys, Add-ons, Storage (disk use, clean-up, where the add-ons are kept), General (theme, language, graphics card, window), Privacy & data (data folder, error reports, support info, export or delete your account) and About (updates, licenses).

• Add-ons are optional downloads, each with a "why" and a "skip it if":
  • Live voice engines: Instant voice cloning (powers Speak, Dub, file conversion and instant cloning; Mimicore picks the model size for your PC) and Trained voices (RVC) (0.3 GB, runs voices live: trained and imported voices and gallery voices marked live + offline). "Advanced: model tiers" lists the Fast, Balanced and Ultra model files for people who want to manage them.
  • Voice training (2.5 GB, NVIDIA only) lets this PC train RVC voices. No NVIDIA card? Use free cloud training instead.
  • Video dubbing (about 3 GB): translates a video's speech into your voice in another language, offline.
  • Cloud training (optional): connect your Google Drive to train on a free Google GPU.
• Storage: shows how much disk space Mimicore uses (add-on models, the Voice Training pack, voices, training runs, projects, results, temporary files, logs) and how much is free. Clean up removes only what Mimicore can make again: temporary files older than a day, leftovers of stopped downloads, and old results (older than 30 days, 90 days or a year) that no saved project uses; it asks before deleting results. Voices, recordings, unfinished trainings (checkpoints), saved projects and installed add-ons are never touched. Where the add-ons are stored is by default inside the data folder (%LOCALAPPDATA%\Mimicore\models); Move to another folder… copies the add-on models and the Voice Training pack (about 2 GB and several GB) to a folder on another drive inside this PC (not a network or removable drive, not Windows or Program Files), checks every file, and after you restart Mimicore uses the new place and removes the old copy. Cancel leaves everything as it was. The Mimicore uninstaller does not delete a custom storage folder: delete it yourself if you remove Mimicore for good. If that drive is missing when Mimicore starts, downloads pause and the page offers to choose another folder or use the default one (the add-ons would then be downloaded again). Mimicore warns before a download or a training when the drive is too full.
• Account: your plan (Free or Studio, with a "trial" tag during a free trial), See plans, sign out.
  • Free plan: all 28 effects and the soundboard, unlimited live AI voice with the 19 gallery voices, 60 minutes a day with your own voices, one voice clone, Speak clips up to 3 minutes (about 2,700 characters), Dub and conversion clips up to 2 minutes; no mixing, importing or training.
  • Always free: effects, the soundboard, the gallery and built-in voices.
  • Studio (monthly, annual or lifetime): everything unlimited, including your own voices live, cloning, mixing, importing, training, Speak and Dub.
• Hotkeys: see section 9.
• Models & add-ons: the two live engines, Voice Training and Dubbing. Mimicore measures your PC by itself in the background.
• Appearance & language: dark, light or match Windows. The interface is available in English, हिन्दी, Español, Português, Français, Deutsch, Русский, 日本語 and Bahasa Indonesia.
• Updates: Mimicore checks automatically and can download updates in the background. You choose when: Restart and install now, or Install when I quit (the update installs after you close Mimicore, never while Live or a job is running, and Mimicore does not reopen).
• Downloads: an interrupted add-on or model download continues where it stopped (also after a restart or a lost connection) and every file is checked against its known checksum at the end. The progress shows speed and time left; after a failure it retries with growing pauses. On a slow or offline PC, download the add-on files on another PC (or earlier) and use Install from a folder... or Install from files... on the add-on in Settings > Add-ons; Mimicore checks every file the same way and tells you what is still missing. Mimicore uses the Windows proxy setting or the HTTPS_PROXY variable; if a proxy blocks a download it says so.
• Taskbar and notifications: the Mimicore button on the Windows taskbar shows the progress of downloads, training and renders. When a voice training or a long render (20 seconds or more) finishes while Mimicore is in the background, Windows shows a notification and the button flashes. Switch it off in Settings > General > Notify me when long jobs finish.
• Processing: pick the graphics card. Any DirectX 12 card works: NVIDIA, AMD or Intel.
• Cloud training: see which Google account is connected, or disconnect it.
• Your data: open the data folder, and optionally Send error reports (off by default). Performance diagnostics (technical data only: timings, error counts and short redacted log lines, app and Windows versions, the graphics card name, power source and battery level, and the installation's random PC id; linked to your account while you are signed in) are on by default and can be switched off under Settings > Privacy & data. Audio, recordings, scripts, file and voice names are never included. Local testing ZIPs are separate.

Gaming with friends on Discord

1. Set up the virtual mic (section 2) and select CABLE Output in Discord.
2. Pick a voice; adjust the trained-voice latency and quality slider before pressing Start.
3. Set a Push-to-talk key and a Start / stop hotkey.
4. Add soundboard clips with hotkeys for instant reactions.

Streaming and VTubing with OBS

1. In OBS, add Audio Input Capture › CABLE Output.
2. Bind Switch straight to a voice hotkeys for each character, plus Mute.
3. Use Add an effect (Radio, Cave, Robot) for scenes, and the soundboard for alerts and memes.

Localise your YouTube videos and Shorts

1. Clone your own voice (an instant voice, or train one for the closest likeness).
2. Dub a video into Hindi, Spanish, Portuguese and more, all in your own voice.
3. Replace the audio track in your editor and upload the SRT as subtitles.

Voiceovers, narration and audiobooks

1. Optional: for the most polished speech install the Ultra engine size (Settings › Add-ons › Advanced).
2. In Speak (or Audiobook Studio for a book), paste or import your script, pick your voice and language, and choose Save to file.
3. Long texts are rendered sentence by sentence, chapter by chapter in Audiobook Studio, and an interrupted render resumes.

Private, anonymous calls

1. Pick a gallery voice or a mix, so the voice belongs to no real person.
2. Select CABLE Output in Zoom, Teams or Discord.
3. Turn on noise suppression and a Light noise gate for a clean sound.

Role-play, D&D and characters

1. Create characters with the mixer and effects (Giant, Ghost, Elf, Dark Lord…).
2. Bind one hotkey per character and switch mid-sentence.

Your own voice, typed

For people who can't speak at times (illness, a sore throat, quiet places): record an instant voice while you can. Speak then reads typed text aloud in your own voice, live into a call.

Speak another language in your own voice

Type in Spanish, Japanese or Hindi in Speak. Your cloned voice says it with natural pronunciation, which is great for presentations and practice.

The Support tab (sign in first) is an assistant that knows the whole app and website. Ask anything: setup, voices, plans, errors.

• If it can solve it, it answers with the exact steps and a link to the guide.
• If it can't, it creates a support ticket for you (Talk to a person does this on request). You get an email at every step: received, confirmed as a real defect, fix in progress, a test build ready, resolved.
• Report a bug: type what you did and roughly when it happened, then press Report a bug. Mimicore picks the matching part of its local log (it records the pages and buttons you used, never what you type), removes your Windows user name and email addresses, and shows you what it would attach. Nothing is sent until you press Send ticket with log (or Send without log).
• Your tickets are listed at the top with their status. When a fix is ready you are emailed a link to an early build, or open Settings › Updates › Check for early release. Early builds are never installed automatically. Try what went wrong and press It is fixed or Still broken.
• Send feedback: for an idea, praise or a question (not a problem you need solved), press Send feedback on the Support page or in Settings › About, or use the form at mimicore.app/feedback without an account. Pick what it is about, add a rating if you like, write your message and press Send. Leaving an e-mail address (and ticking the box that lets us write back) is optional, as is Attach technical details (your app and Windows version, graphics card type and, while Live runs, its speed numbers; never audio, file names or logs). Offline, the app keeps your message and sends it when you are back online. Feedback is kept 12 months.
• Check my setup: if something does not work, press Check my setup on the Support page. It runs quick, read-only checks (a few seconds, all at once) of Windows and WebView2, Windows microphone privacy, your microphone (and whether sound arrives), headphones or speakers, the virtual cable, the graphics card, the add-ons (size and SHA-256 against the list that ships with the app), free disk space, the internet, your Mimicore version and your plan. Each result is green, amber or red with a plain explanation and, where one exists, a button that opens the right Windows settings or Mimicore page. Nothing is changed, downloaded or started, and an offline PC is fine for Live. Copy report gives a text with no file paths, e-mail addresses or names; Send with feedback opens Send feedback with that report in the message. Not installed yet? The page mimicore.app/requirements makes a guess from your browser (nothing leaves it); the app's own check is the authoritative one.
• Deleting your account deletes your tickets, chats, logs and feedback.

• Your voice, your device, your cloud. Never ours. Live voice, cloning, effects, speech, dubbing and training on your PC all run locally. Microphone audio is processed in memory and never stored or sent.
• Optional cloud training uses only your own Google Drive and Colab. Mimicore can see only the files it creates there.
• The Support assistant sends what you type to our server and an AI provider; do not type passwords or payment details. The same applies to the optional AI script polish and random narration in Audiobook Studio, which send the script to an AI provider through our server when you are signed in. Tickets (conversation plus a log you approve) are kept until you delete your account. We otherwise store only your account email and plan. Error reports are off unless you turn them on.
• Clone only your own voice or voices you have permission to use. Never use a cloned voice to deceive, defraud or impersonate anyone. Generated files are marked as AI-generated.
• Full texts: Acceptable Use · License · Privacy.

Open Settings > Privacy & data > Testing log. Start a session, test the app, and press Mark issue immediately when something goes wrong. The session continues after restarting Mimicore. Finish with Stop testing log and Export testing log. Review the ZIP before sharing it; no automatic upload occurs. It contains system details, app events, errors and detailed Live timings, but no audio attachments or typed scripts. Device and file names can remain after redaction. The local folder is %LOCALAPPDATA%\Mimicore\testing-logs (or your MIMICORE_HOME). Each session stops at 128 MB, with four sessions retained; move exported older files elsewhere to make room. Detailed logging adds some overhead.