Mimicore™
☰

What's new

Mimicore updates itself (the Microsoft Store version updates through the Store). Here is what changed, newest first. Full technical notes are on GitHub.

2.9.5

Mimicore has a new home: mimicore.app

  • Changed: the website, account page and share links now use mimicore.app. The old mimicore.pages.dev address keeps redirecting, so existing links still work.

2.9.4

Live starts on slower graphics cards, VB-CABLE installs reliably

  • Fixed: on a slower graphics card (for example Intel UHD), the first Live start of a trained voice could be cancelled after 2 minutes. Live now waits up to 20 minutes while the voice is prepared, and the message no longer blames your internet connection.
  • Fixed: installing the virtual microphone (VB-CABLE) could fail with “pack.zip does not exist” when an antivirus removed the download. It is now unpacked directly from memory.

2.9.3

Importing a trained voice works again

  • Fixed: a trained voice exported from Mimicore (a .mimivoice file) could not be imported again. It was refused with “A data file inside this voice file is not in the expected format”. Trained voices now import normally, on the same PC or another one.

2.9.2

Instant voices live on any DirectX 12 graphics card

  • Live: instant and gallery voices now run through the new Instant live voices add-on (free, Settings > Add-ons, about 500 MB). It runs inside Mimicore on any DirectX 12 graphics card, sounds much better than the old instant live engine and plays about 1 second behind you. Trained voices stay the lowest-delay choice. Without the add-on, instant voices work in Speak, Dub and file conversion.
  • Voices are named by what they are: Instant (cloned from a short recording, also the gallery voices and mixes), Trained (speech) and Trained (singing). The Voices page has a tab for each.
  • Trained voices: when your PC cannot keep up, Live offers a Try N ms blocks button next to the notice. Nothing changes until you press it.
  • A Live start that fails no longer uses one of your free minutes: a minute counts only once your voice is running. A graphics card too slow for Instant live voices is told at once, and with Settings > General > Run on set to CPU only, Live explains why instant voices are not listed.
  • Older graphics cards: Settings > Add-ons has a Compatible pack for cards that cannot run the standard voice engine.
  • Large add-on downloads no longer freeze the app, and the add-ons list updates as soon as an add-on is installed or removed.
  • Mimicore now loads its model and runtime files only from its own folders.

2.9.1

Smoother Live on laptops, and clearer diagnostics

  • Live: on laptops and slower graphics cards the graphics card is now also woken when you start talking after a long silence, so the first words come out on time. Until now this only happened on faster cards.
  • While Live runs, Windows no longer slows Mimicore down to save power, for example with the window minimized during a call.
  • Starting Live no longer fails once in a while with a Windows audio thread error: Mimicore now retries the microphone shortly after it.
  • Performance diagnostics now include the exact graphics card, the power source, battery level and Battery saver, and are linked to your account while you are signed in. You can switch them off under Settings > Privacy & data.
  • A laptop running on a low battery slows the graphics card down a lot. If Live glitches on battery, plug in the charger.
Early builds 2.9.0-beta.43 to 2.9.1-beta.1 (6 releases for testers)

2.9.1-beta.1early build

Laptops: on time after a pause, and battery details in diagnostics

  • Live: on laptops and slower graphics cards the graphics card is now also woken when you start talking after a long silence, so the first words come out on time. Until now this only happened on faster cards.
  • While Live runs, Windows no longer slows Mimicore down to save power, for example with the window minimized during a call.
  • Performance diagnostics now include the exact graphics card, the power source, battery level and Battery saver, and are linked to your account while you are signed in, in every version. A laptop low on battery slows its graphics card down a lot, and this shows it. Settings > Privacy & data shows exactly what is sent and turns it off.

2.9.0-beta.47early build

Measuring the first moment after a long pause, and cleaner diagnostics

  • The Live log now records how long the first block of speech took after a long silence, next to the normal block time, so the wake-on-speech improvement can be checked on any PC.
  • Diagnostics samples report only the underruns since the previous sample instead of repeating the whole session's count, and a failed update check caused by the network is no longer counted as an error.

2.9.0-beta.46early build

Cleaner logs after a cancelled start

  • Pressing Stop while a voice is still loading is no longer recorded as an error, so the error counts in the diagnostics only show real problems.
  • The Live session summary no longer shows a wrong uptime for a start that was cancelled or failed while loading.

2.9.0-beta.45early build

No more stall when you start talking after a long pause

  • Live: the first moment of speech after a long silence no longer takes about four times longer than normal. The graphics card is now woken just as you start to speak, so the first block comes out at normal speed (checked on a graphics card at the 640 ms and 1.2 s positions; the shortest positions improve less).
  • Fixes from a screen-by-screen review: duplicate labels in the voice file export and import windows, a clipped avatar and plan label in the narrow sidebar, tour tips that could slide off the top of a small window, and two labels that stayed in English in other languages.
  • The Send feedback window keeps its Send button in view when a message is long or an error is shown.

2.9.0-beta.44early build

Your name in the sidebar and an optional name field

  • The account card at the bottom of the sidebar is easier to read: a round avatar with your initials, your name, and your e-mail beneath it, then your plan.
  • Settings > Account has an optional Your name field, mainly for e-mail sign-ins that have no name. It is used to greet you in e-mails from us, can be cleared at any time, and a name you set is never overwritten by Google or Discord.

2.9.0-beta.43early build

Your account in the sidebar, plainer model names and a diagnostics switch

  • The bottom of the sidebar shows Sign in, or your account above your plan, and opens Settings > Account.
  • Trained voices are named by what they are for, speech or singing, everywhere, and About lists the models Mimicore uses. The Add-ons page no longer shows quality-level names next to Instant voice cloning; the extra levels are under Advanced.
  • Anonymous performance diagnostics (technical numbers only: no audio, recordings, file names or personal data) are now on by default in every version, with a clearly labelled switch and a Reset PC id button in Settings > Privacy & data. There is no pop-up. Turning it off stops sending and discards what was waiting.
  • Fixes: error messages that pointed to a Settings page that does not exist, and a few labels that stayed in English in other languages.

2.9.0

Saved projects, a Latency and quality slider, Audiobook Studio and accent matching

  • Live: trained voices get one Latency and quality slider with 15 positions, from about 80 ms to 1.2 s of buffering. It replaces the old Low, Medium and High buttons and the automatic quality system, and your choice stays fixed until you stop Live. Keep it toward the left for calls and games; the right-hand end is for streams that prefer smoothness over speed. Instant voices and effects have no slider.
  • Steadier Live on slower or busier graphics cards: a larger output buffer on the longer positions, a small keep-alive job that stops the card from slowing down between blocks, a short notice with what to try when the card is too busy, and an optional "Faster, slightly less accurate" setting (off by default) that searches a smaller part of the voice's training data. Tested by measurement and simulation on one graphics card; how much it helps on yours still depends on the PC.
  • Saved projects: Speak scripts, titles and voice settings are saved on your PC as you type and survive closing the app. In Audiobook Studio you can import PDF, Word or scanned pages, split a book into chapters and render them; an interrupted render resumes from the last verified chapter.
  • Guided setup: choose Live, voiceover or Dub and Mimicore downloads only what that needs, and finishes after you have a real result. No background performance test runs any more.
  • Accent matching: on a trained voice, add a short reference recording and Speak, Dub and Convert carry part of that accent (15 regional options). It is not an exact copy, and live voice keeps your own accent.
  • Voice training asks what the voice is for: speech is included and the singing model is an optional download. One training on this PC and one in the cloud can run side by side, with a progress card while you use the app.
  • Speak has a Speed slider, saves your text as you type and can undo script polishing. Word imports and overlapping jobs recover more safely.
  • Safer: your sign-in and Google Drive details are protected by Windows on this PC, updates wait for active processing, the app can only open files you choose, and Mimicore retries when Windows is slow to start its web view.
  • Support: the in-app assistant explains the new controls, and a ticket you create gets an email at every step, including an early build to try when a fix is ready.
  • Looking after your PC: Settings > Storage shows what Mimicore uses and cleans up safely (never your voices, recordings or unfinished trainings) and lets you choose where add-ons live; downloads resume and are verified; "Install when I quit" applies an update when you close the app; the taskbar and a Windows notification tell you when a long training or render finishes; Support > Check my setup tests your Windows, microphone, output, virtual cable, graphics card and connection; after an unexpected exit Mimicore offers a calm recovery and a safe mode; double-clicking a .mimivoice file opens the import dialog; and the new window comes to the front after Restart and install.
  • Not in this release: Mimicore Web is still closed, and international card payments are not enabled yet.
Early builds 2.9.0-beta.29 to 2.9.0-beta.42 (14 releases for testers)

2.9.0-beta.42early build

The Faster option is a switch, and a safer upgrade from older versions

  • Faster, slightly less accurate under the Latency and quality slider is now a proper on/off switch (Space or Enter toggles it, it is inert while Live runs). It is still off by default, never turns on by itself, and is remembered in saved set-ups.
  • The first time this version starts after an older one, a copy of your previous settings (without your sign-in) is kept as settings.before-2.9.json in the Mimicore data folder, in case you ever want to go back.

2.9.0-beta.41early build

Clearer Live logs and diagnostics that are no longer lost

  • The Live log now says how many index rows are actually searched, the cap that applied and how many rows the voice's index holds, and notes when the cap or the Faster option makes Live search every second or third row.
  • While you are silent the running log line is short (no new blocks) instead of repeating the same numbers.
  • If the server refuses a diagnostics batch for its content, it is now kept for a retry for up to a day (and while you stay on the same version) and the local log says so, instead of being dropped without a trace.

2.9.0-beta.40early build

Tidier Live logs and a safer audio start-up

  • Live's session summary no longer counts a brief dry spell in the first two seconds, while the output buffer settles, as an audible gap; gaps and late blocks now agree.
  • The running summary line is written once a minute even after the PC sleeps, instead of in a burst.
  • Live now claims the multithreaded Windows audio apartment first on its audio thread, a defensive step for the laptop start-up error "Cannot change thread mode after it is set". It is not proven to fix that error.
  • A GPU keep-alive that failed is no longer reused by the next Start.
  • Layout polish: the support chat box now centres its placeholder and text, the Speak title field matches the other fields, long headings such as "Text to speech" no longer get squeezed in German, Russian or Hindi, the narrow sidebar shows the free-minutes pill on two lines, and "Import folder…" is now translated.
  • More layout polish: Settings tabs no longer break letter by letter in Japanese (they wrap onto a second row), the narrow window shows a compact "?" for Guided setup, and a few labels that stayed English in other languages are translated (Accent / region in the training dialog, download sizes, the Run on option, the word count under Speak).
  • The in-app assistant and the Help page now explain the Faster option and the busy-graphics-card notice, and no longer mention the old Low, Medium and High buttons.

2.9.0-beta.39early build

An optional faster mode, and calmer, clearer diagnostics

  • New option under the Latency and quality slider for trained voices: Faster, slightly less accurate. It is off by default and never turns on by itself. It matches your voice against a smaller part of its training data, so each block finishes sooner on a slow or busy graphics card. Turn it off to get exactly the previous behaviour; it applies the next time you press Start.
  • The busy-graphics-card notice is now much less eager: it needs about six late blocks within two minutes, or processing that cannot keep up with the block length for about a minute. It shows at most once every 15 minutes and goes away by itself.
  • Live now writes a running summary line every minute, so a session that is closed or crashes still leaves its totals. The summary and the diagnostics also record the slider position, whether it changed since the last session and whether the faster option was on.
  • Diagnostics no longer repeat identical timings while you are silent, which also makes them a little smaller.

2.9.0-beta.38early build

Steadier Live on busy or slower graphics cards

  • On positions above 200 ms, when blocks keep arriving late, Live now lets its output buffer grow past the old fixed limit (up to nine tenths of a block) and no longer trims the safety margin away right after a slow block. A busy laptop graphics card now gets a steadier, slightly longer delay instead of a cut every few seconds.
  • The GPU keep-alive pauses while Live's own processing already keeps the graphics card busy (about 45% of the time or more), so it no longer adds load to slower graphics cards.
  • When the graphics card stays too busy, Live shows a short notice with what to try: plug in a laptop's charger, and close or limit games and apps that use the graphics card.
  • A harmless glitch or re-route notice from a speaker or virtual mic no longer restarts playback (the microphone got the same fix in beta 29).
  • Live logs now record the output sample rates, whether the PC runs on battery and how long the keep-alive held back.

2.9.0-beta.37early build

Fewer early glitches on the longest positions, and clearer glitch logs

  • On the 480 ms and longer positions Live now starts with a larger output buffer for the first 20 seconds, when processing is slowest, and gives the extra back within about a minute instead of keeping it for the whole session.
  • The log line written when Live stops now lists the first late blocks (when they happened, the buffer at that moment and how long the gap was). A late block heard on two outputs is counted once in that list; the glitch total still counts each output.

2.9.0-beta.36early build

Steadier first seconds of a Live session

  • Live now learns how much processing time varies after 4 blocks instead of 8, and a new session starts from what the previous session of a similar block length measured. This targets the occasional glitch in the first seconds of a session.
  • On the first Start after opening Mimicore, Live waits up to a few seconds for the GPU keep-alive to be ready before it begins converting.

2.9.0-beta.35early build

Lighter diagnostics and a smarter GPU keep-alive

  • Beta diagnostics now send far less data: Live snapshots every 15 seconds instead of 5, no duplicate start records, no repeated status lines and shorter timestamps. An ordinary two minutes is now under about 9 KB instead of 20-28 KB, so a long testing day stays inside the daily allowance.
  • The GPU keep-alive added in beta 34 now waits while a real block is being converted, pauses during silence, reuses its setup between starts, no longer delays Start, and is skipped on integrated graphics.

2.9.0-beta.34early build

Faster, steadier Live on the longer positions

  • Testing showed the graphics card slows itself down in the quiet gaps between long blocks. Trained voices on the 160 ms and longer positions now keep it awake with a very small recurring job. On the test PC the processing time per block fell by roughly 25-45% and became much steadier. It uses a little extra graphics memory and a steady share of GPU time while Live runs.
  • The 80 ms position is unchanged. If you prefer the card to rest, start Mimicore with the environment variable MIMICORE_NO_KEEPALIVE=1.

2.9.0-beta.33early build

Fewer dropouts on the longer Latency and quality positions

  • On the positions above about 200 ms, Live now keeps a larger output buffer that grows with the block length and with how much processing time varies, instead of a fixed 40 ms. The longest positions add that buffer as extra delay. Positions up to about 200 ms behave as before.
  • The log line written when Live stops now also reports the length of each audio gap, so remaining dropouts can be traced. Whether this removes your dropouts still needs testing on your own PC.

2.9.0-beta.32early build

Diagnostics keep arriving on busy testing days

  • The daily beta diagnostics allowance is much larger (2 MB per account instead of 256 KB), because one day of heavy testing used it all up and later sessions were refused.
  • When the allowance is reached the app now keeps the report and tries again later, instead of discarding it.

2.9.0-beta.31early build

Beta diagnostics no longer need a sign-in

  • Windows beta diagnostics are sent whether or not you are signed in, so testing from a signed-out PC or an expired session still reaches us. Signed-out reports carry only a random PC ID and cannot be linked to an account.
  • If an upload cannot complete, the app now records why and keeps retrying. Audio, scripts, credentials and private file or voice names are still never sent.

2.9.0-beta.30early build

A wider Latency and quality range

  • The Latency and quality slider now has 15 positions from about 80 ms to 1.2 s of buffering, in 80 ms steps. The old 40 ms end is gone and the new right-hand end is for PCs or streams that prefer smoothness and quality over speed.
  • Settings and saved set-ups from earlier betas move to the nearest new position (40 ms becomes 80 ms, 120 ms becomes 160 ms). Your voices, devices and pitch settings are untouched.
  • Larger positions add delay you will hear. For calls and games, keep the slider toward the left. Listening, hardware and game/call performance still need testing on your own PC.

2.9.0-beta.29early build

Simpler Live tuning and safer startup

  • One 17-position Latency and quality slider sits inside the trained-voice card. It replaces the separate mode and block-size menus; your choice stays fixed until you stop Live and adjust it.
  • Recoverable microphone timing errors no longer force a reconnect. Repeated warnings are limited so useful diagnostics remain available.
  • MC-00097: startup retries a busy WebView2 runtime, then shows a clear error if the window still cannot open. Existing browser data, voices, projects and recordings are preserved.
  • A separate Windows audio-client startup error reported on one laptop remains under investigation. Listening, hardware and game/call performance still need testing on your own PC.
Older releases (8 more)

2.8.5

Speak and Dub finish what they start

  • Speak and Dub could sometimes stop after the first words and leave a long silence. A sentence that does not finish is now generated again automatically.

2.8.4

Stay signed in, and hear about updates

  • An in-app update bar tells people on older versions when a new build is ready.
  • Announcements can now target a plan or a country.
  • You stay signed in until you sign out yourself.
  • Studio is free outside India (Early Bird Gift).

2.8.3 early buildearly build

A more compact app (historical controls)

  • This release used an automatic quality system. Later Windows betas replace it with a fixed latency and quality slider for trained voices.
  • The automatic controls described by this historical release have since been retired in beta development.
  • A tighter layout: the Live page fits on one screen and the Start button is always visible, even in a small window. Pages are denser and the app does less background work when idle or minimised.
  • Live delay at full quality is about 0.7 s for trained voices and 0.5 s for instant voices.
  • Trained voices now come out as loud as your own voice; they were far too quiet before. Output keeps 1 dB of headroom so it cannot distort.
  • Russian text to speech works, and Japanese speech no longer fails. Settings are saved safely, so a power cut while saving cannot wipe them, and the built-in soundboard clips play at an even volume.
  • A redesigned Support page: one chat with a round arrow send button, and a single Create a support ticket option when the assistant cannot solve it. We reply to tickets by email.
  • The Live engine panel opens by itself when you go live and shows the main load (GPU for trained voices) first. When the window is maximised the content stays in a centred column.
  • Dubbing installs its add-on right on the Dub page, and the Add-ons page says which instant-voice sizes are installed.
  • Two plans: Free and Studio. Every new account starts with 30 days of Studio free, no card. After that, Free keeps all 28 effects, the soundboard, hotkeys, unlimited live AI voice with the 19 gallery voices, 60 minutes a day with your own voice, one voice clone, Speak clips up to 3 minutes and Dub and file conversion clips up to 2 minutes. Studio adds unlimited own-voice use, unlimited cloning, mixing and training, and Speak, file conversion and Dub with no clip limit.

2.8.2

Use Mimicore on all your PCs

  • One licence now works on all your PCs, with one PC active at a time and a 5-minute grace period when you switch.
  • Clearer wording about the PC limit.

2.8.1

Gentler prompts and invites

  • A quiet in-app note when your free live minutes run low.
  • An invite card after a good session: share your link and friends get extra free Studio days.
  • Small fixes.

2.8.0

Free Studio days for new accounts

  • New accounts get free Studio days with no card (now 30 days).
  • Invite friends for bonus Studio days.
  • Reminders before a pass or trial ends.
  • Clearer trial and gift wording in all 8 languages.

2.7.x

Auto-tuned live voice

  • Mimicore measured your PC after install and picked the best smooth setting for instant and trained voices (since simplified in 2.8.3).
  • A Live engine panel shows CPU, GPU, block time and glitches while you are live.
  • Add-ons simplified to Instant voice cloning and Trained voices; Voice training and Dubbing stay.

2.6.5 to 2.6.7

Simpler start

  • No consent screens or spoken statements inside the app: the terms are accepted once on the download page.
  • New voices need just a name and a recording.

Want it first? In the app, open Settings, About, Check for early release. Questions? See the FAQ or the contact page.