Bunyi for Mac 1.3.3

Bunyi 1.3.3 — Mac

More room to write

Script, style instruction and voice description editors can now be resized with a drag handle. Multiline instructions make longer prompts easier to read and edit.

Speech fixes

Newline-separated paragraphs retain their pauses during generation.

Security and reliability

• Model and transcription downloads verify pinned digests; the built-in mirror requires a complete checksum manifest.
• Backup and restore reject unsafe file trees and backups larger than the available free space.
• Saved voice identifiers and model paths receive stricter validation.
• Logs redact sensitive download details and omit reference transcripts.
• The native Mac CLI server limits connections and message-read time, and keeps accepting clients after a bad connection or resource shortage.

Updating

Models, saved voices, recordings and settings are retained. This release includes the native Mac CLI archive alongside the app downloads.

Full changes

---

Signed with a Developer ID certificate, notarized by Apple, and stapled.

Requires macOS 15 or later on an Apple silicon Mac.

Open the .dmg and drag Bunyi to Applications. The .zip is the same app for anyone scripting an install.

The first generation in each mode downloads that mode’s model (about 1.5–4.5 GB) with a progress bar; after that it runs offline.

SHA-256:

d307717fe9269d7f7f66cf88be4e443e05b19c1167fd66f4dd4ca77c7d7b0786  Bunyi-1.3.3-8.dmg
7c1eb76433a670b6180519f37d2cb52301cde9265a02587c5aa335e9dd55539d  Bunyi-1.3.3-8.zip
48a038d0e12ed87757586bf44dea97eb679a5df91b583023eeae83a2818a4486  Bunyi-CLI-1.3.3-8-macos-arm64.zip

Bunyi for Windows and Linux 1.3.4

Features

• feat(ui): add resizable multiline script and voice editors (#268)

Fixes

• fix(cli): keep the server accepting after a client drops mid-connect (#291)
• fix(security): apply openrouter security review findings (#290)
• fix(security): harden the macOS server, restore, logging, downloads and site (#283)
• fix(security): harden the Windows server, parser limits and release action pins (#281)
• fix: preserve pauses between newline-separated paragraphs (#267)

Other changes

• chore(dotnet): release 1.3.4
• Security review (codex): findings and fixes (#279)
• Security review (claude): findings and fixes (#278)
• Security review (codex): findings and fixes (#276)

Full changes: dotnet-v1.3.3...dotnet-v1.3.4

---

Installing

Windows — run Bunyi-1.3.4-win-x64-setup.exe, then open Bunyi from Start.
Windows with NVIDIA acceleration — use Bunyi-1.3.4-win-x64-cuda-setup.exe. It replaces the standard edition and retains your data; run the standard installer to switch back. Compatible CUDA and cuDNN libraries must be installed separately.
Ubuntu/Debian — install bunyi_1.3.4_amd64.deb with your package installer.
Fedora — install bunyi-1.3.4-1.x86_64.rpm with your package installer.
Linux packages add Bunyi to your application menu and install native dependencies.

Upgrades and uninstall retain your models, voices, recordings and settings. Close Bunyi before upgrading. Download a newer installer to update the app.

Portable alternatives — unzip the Windows ZIP and run Bunyi.App.exe, or extract the Linux tarball and run ./Bunyi.App.

CLI for agents and terminals — use the separate Bunyi-CLI- archive for your OS, extract the entire folder, and run bunyi --help (Windows: bunyi.exe --help). Add that folder to PATH if desired. No desktop session or .NET installation is required. See the CLI guide for one-shot generation, the resident server, and download progress.

Self-contained: there is no .NET runtime to install. Models download through Bunyi after installation.

The -cuda downloads are the same app built against ONNX Runtime with its GPU package, measured 1.7x faster on long text. Take one only if you have an NVIDIA card and the NVIDIA CUDA Toolkit installed: the provider links against cuBLAS and cudart, which are not shipped here. If CUDA cannot load, the app falls back to the CPU, says so in the log, and generates normally. Everyone else wants the plain download.

Code signing policy

Windows — app/CLI binaries and installers are Authenticode-signed with a Certum certificate and timestamped. The publisher is Shazron Elhazar Abdullah. SmartScreen may still warn on new downloads; check the publisher under More info. SHA-256 checksums are below.

Linux — DEB, RPM and portable downloads are unsigned. Verify them against the SHA-256 checksums published below.

macOS is a separate download, signed with an Apple Developer ID certificate and notarized by Apple.

Team roles — committers, reviewers and approvers: Shazron Abdullah (@shazron).

This program does not transfer any information to other networked systems unless you ask it to: models are downloaded when you first use a mode, and nothing else leaves your machine.

What works

All three modes — preset voice, voice design and voice clone — plus saved voices, history, and backup and restore.

The first generation in each mode downloads the model for that mode (3.9–5.9 GB) with a progress bar and a time estimate; after that it runs offline. Doctor checks there is room before the download starts.

Requirements

A 64-bit Windows or Linux machine. Generation runs on the CPU unless you took a -cuda download and have the toolkit for it; long text can reach about 7 GB of memory on preset voice, so 8 GB is tight and 16 GB is comfortable.

SHA-256:

19130b852105e77e01466e4c958f2a5d15eb661357185011c228c6551db250f5  Bunyi-1.3.4-linux-x64-cuda.tar.gz
bf64baed89a49e40ef92e3e8bba14ee0c8717735e8f978ce198f2935cb350783  Bunyi-1.3.4-linux-x64.tar.gz
3d966542db178427e8245fe831bccd6d7f1d666fc694c4b98cf41a77a26c0d2e  Bunyi-1.3.4-win-x64-cuda-setup.exe
9b19e6e9737d429dcc55e384ceac5083fbf2db9eddc79e2fe91a60db4d48d2ab  Bunyi-1.3.4-win-x64-cuda.zip
eaec8cfbd2492056bf2f39c737da94d28d58ee7fd9560cc673bb3c6768b73246  Bunyi-1.3.4-win-x64-setup.exe
90e9d596919b88e62e1dad7ed67a48208e015d4d20b0adac80fca74330584832  Bunyi-1.3.4-win-x64.zip
8832cc494977ff97cf51c24b9a0372feb385b64a67b82f0529e5ae317a95e9d9  Bunyi-CLI-1.3.4-linux-x64-cuda.tar.gz
3cbc02bf9d0e1e6f4bf21af44727aad5050e2c69018e3634e804ff7059cac10d  Bunyi-CLI-1.3.4-linux-x64.tar.gz
f0cb8290af9d84030aa46d10a3eeeb54d196700f69d2f2c9a5771a00fc499d0c  Bunyi-CLI-1.3.4-win-x64-cuda.zip
c77a448afb9ba9ac84e3716587b83453ea56945ed5619ef40c2e811162528e84  Bunyi-CLI-1.3.4-win-x64.zip
69c53d34b6a2984bdb2fa562dfe4dadccab3eb8e88089314af66e34e6e4624ca  bunyi-1.3.4-1.x86_64.rpm
93e6af01cb6ea1a3ceb315d9603422f57533f2da5b1558577063d504a4189554  bunyi_1.3.4_amd64.deb

Bunyi for Mac 1.3.2

Bunyi 1.3.2 — Mac

Smoother long-text speech

• Adjust the pause between generated sections. Settings → General → Sentence gap (ms) defaults to 300 ms and accepts 0–5000 ms. Zero disables added silence. Changes apply to the next recording; pauses within a section remain controlled by the voice model.
• Voice Clone automatic references end at a quiet pause. Longer reference recordings are sampled up to ten seconds without cutting through speech. Transcription and cloning use the same selection, and saved voices retain it. If no suitable pause is found, Bunyi asks for a shorter recording.
• History stays locked during generation. Clicks and keyboard navigation keep the active generation tab visible until work finishes or stops.

More reliable downloads

Temporary model-server failures retry with backoff. Explicitly paused downloads still stop promptly, and Stop remains available while waiting.

Updating

Models, saved voices, recordings and settings are retained. To apply the reference-boundary fix to an existing clone, clear its old transcript and generate again so Bunyi selects and transcribes a fresh reference window. Existing recordings are unchanged.

The native Mac CLI receives the same sentence-gap default, automatic reference selection and saved-reference alignment fixes.

---

Signed with a Developer ID certificate, notarized by Apple, and stapled.

Requires macOS 15 or later on an Apple silicon Mac.

Open the .dmg and drag Bunyi to Applications. The .zip is the same app for anyone scripting an install.

The first generation in each mode downloads that mode’s model (about 1.5–4.5 GB) with a progress bar; after that it runs offline.

SHA-256:

7ace7538708238ff438efbade7f83f8eaaa0b231e5b05c69f5be4288a820605a  Bunyi-1.3.2-7.dmg
ec2402e79e6103f38993126be357f9ec5c0c63cb4b7e9d560c74295d005530a3  Bunyi-1.3.2-7.zip
11d79581ac0dc638e3f4f2ecb25a31890a71f854bd2c9650cc073aaf19135ec4  Bunyi-CLI-1.3.2-7-macos-arm64.zip

Bunyi for Windows and Linux 1.3.3

Bunyi 1.3.3 — Windows and Linux

Smoother long-text speech

• Adjust the pause between generated sections. Settings → General → Sentence gap (ms) defaults to 300 ms and accepts 0–5000 ms. Zero disables added silence. Changes apply to the next recording; pauses within a section remain controlled by the voice model.
• Voice Clone automatic references end at a quiet pause. Longer reference recordings are sampled up to ten seconds without cutting through speech. Transcription and cloning use the same selection, and saved voices retain it. If no suitable pause is found, Bunyi asks for a shorter recording. Desktop and CLI follow the same rule.

Downloads and packaging

• Temporary model-server failures retry with backoff. Explicitly paused downloads still stop promptly, and Stop remains available while waiting.
• Microsoft Store packages include the native runtime libraries required for inference, transcription and audio.
• Standard CPU and NVIDIA CUDA Windows installers retain the existing Certum signing and timestamp verification.

Updating

Close Bunyi and run the installer for your edition. Models, saved voices, recordings and settings are retained. To apply the reference-boundary fix to an existing clone, select its original reference recording, clear the transcript and generate again; save the voice again to keep the new selection. Existing recordings are unchanged.

---

Installing

Windows — run Bunyi-1.3.3-win-x64-setup.exe, then open Bunyi from Start.
Windows with NVIDIA acceleration — use Bunyi-1.3.3-win-x64-cuda-setup.exe. It replaces the standard edition and retains your data; run the standard installer to switch back. Compatible CUDA and cuDNN libraries must be installed separately.
Ubuntu/Debian — install bunyi_1.3.3_amd64.deb with your package installer.
Fedora — install bunyi-1.3.3-1.x86_64.rpm with your package installer.
Linux packages add Bunyi to your application menu and install native dependencies.

Upgrades and uninstall retain your models, voices, recordings and settings. Close Bunyi before upgrading. Download a newer installer to update the app.

Portable alternatives — unzip the Windows ZIP and run Bunyi.App.exe, or extract the Linux tarball and run ./Bunyi.App.

CLI for agents and terminals — use the separate Bunyi-CLI- archive for your OS, extract the entire folder, and run bunyi --help (Windows: bunyi.exe --help). Add that folder to PATH if desired. No desktop session or .NET installation is required. See the CLI guide for one-shot generation, the resident server, and download progress.

Self-contained: there is no .NET runtime to install. Models download through Bunyi after installation.

The -cuda downloads are the same app built against ONNX Runtime with its GPU package, measured 1.7x faster on long text. Take one only if you have an NVIDIA card and the NVIDIA CUDA Toolkit installed: the provider links against cuBLAS and cudart, which are not shipped here. If CUDA cannot load, the app falls back to the CPU, says so in the log, and generates normally. Everyone else wants the plain download.

Code signing policy

Windows — app/CLI binaries and installers are Authenticode-signed with a Certum certificate and timestamped. The publisher is Shazron Elhazar Abdullah. SmartScreen may still warn on new downloads; check the publisher under More info. SHA-256 checksums are below.

Linux — DEB, RPM and portable downloads are unsigned. Verify them against the SHA-256 checksums published below.

macOS is a separate download, signed with an Apple Developer ID certificate and notarized by Apple.

Team roles — committers, reviewers and approvers: Shazron Abdullah (@shazron).

This program does not transfer any information to other networked systems unless you ask it to: models are downloaded when you first use a mode, and nothing else leaves your machine.

What works

All three modes — preset voice, voice design and voice clone — plus saved voices, history, and backup and restore.

The first generation in each mode downloads the model for that mode (3.9–5.9 GB) with a progress bar and a time estimate; after that it runs offline. Doctor checks there is room before the download starts.

Requirements

A 64-bit Windows or Linux machine. Generation runs on the CPU unless you took a -cuda download and have the toolkit for it; long text can reach about 7 GB of memory on preset voice, so 8 GB is tight and 16 GB is comfortable.

SHA-256:

aebf3de363a4698f4e975bbf34ff968f0c0f21714069619a7ce2d470015c3d67  Bunyi-1.3.3-linux-x64-cuda.tar.gz
935f066d76d440acd746d75b29617105982c92b7fbda21dbd8a3844a52c11e36  Bunyi-1.3.3-linux-x64.tar.gz
f80f056afa5e8384b8443bdf41834de30ee1abc2eff42ca0d0d30213e74564c3  Bunyi-1.3.3-win-x64-cuda-setup.exe
0c47ee3ca77d5954a010a21eb3ce27daf6c478200ff896aae19db287afc0298a  Bunyi-1.3.3-win-x64-cuda.zip
46699b65217de8073157f937758b14edba3e0f311d7f3c6c86ddfefa5cb63da4  Bunyi-1.3.3-win-x64-setup.exe
443cd43bf02a02964433bd7bf391a1f4b50ccaa702db5d8f4c598723958b39fa  Bunyi-1.3.3-win-x64.zip
6d90846df0216e61e5694f66dfc5cc84df831922035926314a535cd8c3b722a5  Bunyi-CLI-1.3.3-linux-x64-cuda.tar.gz
0874b56233bac89e4afebbbf250064acfbb3a9758ed1849097ba33f3bc5499c1  Bunyi-CLI-1.3.3-linux-x64.tar.gz
0707e05e5823b8f8137f7a983655b487d13af4cb86c232de8e0588fee1461402  Bunyi-CLI-1.3.3-win-x64-cuda.zip
19eebe7966189fd25106bef4b67dc1739f645703454d550a86f955319f72a6b6  Bunyi-CLI-1.3.3-win-x64.zip
c3c582edc3c9a471eca47712a951d5be0dd79180353e581d9017155d0f926630  bunyi-1.3.3-1.x86_64.rpm
937550c28090ef892c1d1bc78b724755cd43386afe2ad467bea15bd2448d6d65  bunyi_1.3.3_amd64.deb

Bunyi for Windows and Linux 1.3.2

Bunyi 1.3.2 — Windows and Linux

Signed Windows downloads

Windows desktop and CLI binaries, setup installers and uninstallers are now
digitally signed with a Certum certificate issued to Shazron Elhazar Abdullah.
Both standard CPU and NVIDIA CUDA editions are signed and timestamped. The
release process verifies signatures before publishing and generates SHA-256
checksums from the final files.

SmartScreen may still warn on a new download. Use More info to check the
publisher. Signing identifies the publisher and lets Windows detect changes
to the signed files; it does not guarantee an immediate SmartScreen reputation.

Updating

Close Bunyi and run the 1.3.2 installer for your edition. It upgrades the existing
installation while retaining models, voices, recordings and settings. Standard
and CUDA installers can replace each other in the same installation.

Existing 1.3.1 assets and checksums remain available unchanged. This is a
packaging and release update; speech-generation behavior is unchanged.
Linux packages and archives retain checksum verification. Native macOS releases
keep their independent version and existing Apple signing/notarization.

---

Installing

Windows — run Bunyi-1.3.2-win-x64-setup.exe, then open Bunyi from Start.
Windows with NVIDIA acceleration — use Bunyi-1.3.2-win-x64-cuda-setup.exe. It replaces the standard edition and retains your data; run the standard installer to switch back. Compatible CUDA and cuDNN libraries must be installed separately.
Ubuntu/Debian — install bunyi_1.3.2_amd64.deb with your package installer.
Fedora — install bunyi-1.3.2-1.x86_64.rpm with your package installer.
Linux packages add Bunyi to your application menu and install native dependencies.

Upgrades and uninstall retain your models, voices, recordings and settings. Close Bunyi before upgrading. Download a newer installer to update the app.

Portable alternatives — unzip the Windows ZIP and run Bunyi.App.exe, or extract the Linux tarball and run ./Bunyi.App.

CLI for agents and terminals — use the separate Bunyi-CLI- archive for your OS, extract the entire folder, and run bunyi --help (Windows: bunyi.exe --help). Add that folder to PATH if desired. No desktop session or .NET installation is required. See the CLI guide for one-shot generation, the resident server, and download progress.

Self-contained: there is no .NET runtime to install. Models download through Bunyi after installation.

The -cuda downloads are the same app built against ONNX Runtime with its GPU package, measured 1.7x faster on long text. Take one only if you have an NVIDIA card and the NVIDIA CUDA Toolkit installed: the provider links against cuBLAS and cudart, which are not shipped here. If CUDA cannot load, the app falls back to the CPU, says so in the log, and generates normally. Everyone else wants the plain download.

Code signing policy

Windows — app/CLI binaries and installers are Authenticode-signed with a Certum certificate and timestamped. The publisher is Shazron Elhazar Abdullah. SmartScreen may still warn on new downloads; check the publisher under More info. SHA-256 checksums are below.

Linux — DEB, RPM and portable downloads are unsigned. Verify them against the SHA-256 checksums published below.

macOS is a separate download, signed with an Apple Developer ID certificate and notarized by Apple.

Team roles — committers, reviewers and approvers: Shazron Abdullah (@shazron).

This program does not transfer any information to other networked systems unless you ask it to: models are downloaded when you first use a mode, and nothing else leaves your machine.

What works

All three modes — preset voice, voice design and voice clone — plus saved voices, history, and backup and restore.

The first generation in each mode downloads the model for that mode (3.9–5.9 GB) with a progress bar and a time estimate; after that it runs offline. Doctor checks there is room before the download starts.

Requirements

A 64-bit Windows or Linux machine. Generation runs on the CPU unless you took a -cuda download and have the toolkit for it; long text can reach about 7 GB of memory on preset voice, so 8 GB is tight and 16 GB is comfortable.

SHA-256:

2a78acda7f843473240bc0f45987c49d8430b30b5b425685f3bd59e4d076c989  Bunyi-1.3.2-linux-x64-cuda.tar.gz
bd43f566fb3022065320086f0e1c23a0d93e34814bb3de275e7c540d06c23888  Bunyi-1.3.2-linux-x64.tar.gz
8ee13f0c75e34ddcfbb9c2e94c49d4b0238fc3c5e4471a6731cb694e80f79ccd  Bunyi-1.3.2-win-x64-cuda-setup.exe
e858671ee5b208e04ad2ab57c5f509615705cfd65e2c611d695b834c36db6431  Bunyi-1.3.2-win-x64-cuda.zip
41cccbb18a89c93412d6c3a466af592e24be9190d7a0228bebe875b90787b201  Bunyi-1.3.2-win-x64-setup.exe
dd52b4310044bf057c07c29dc80d52dd288cd72f1c621395ffe515189b42343f  Bunyi-1.3.2-win-x64.zip
99dcb6e4ded624d47278ce0a35fdb7678f19fd9186675320733a15544d1e2192  Bunyi-CLI-1.3.2-linux-x64-cuda.tar.gz
6d57a6d2f9254d16b71ec42c09268e339d8f3161a76bf3ee242d8181b81f30f7  Bunyi-CLI-1.3.2-linux-x64.tar.gz
df8d3aebd40ac2fdae295e7a0c279b5c6101bba2f80d253675ffa3fe49b80bea  Bunyi-CLI-1.3.2-win-x64-cuda.zip
7a9b2a76388157ca6ad8497fe16a6665f7bbe203594018eaf62d825184418063  Bunyi-CLI-1.3.2-win-x64.zip
e24035020ac65dacc6b38e425bab1c5662ee8efea8fe429c8c408e2ebe8843a4  bunyi-1.3.2-1.x86_64.rpm
0df44f1bf8e3e2e4dc20fec769ca668133db7c81e60deeaf0a8b05c81a4f11a0  bunyi_1.3.2_amd64.deb

Bunyi for Mac 1.3.1

Features

• feat: default all packages to the Bunyi mirror (#249)

Full changes: v1.3.0...v1.3.1

---

Signed with a Developer ID certificate, notarized by Apple, and stapled.

Requires macOS 15 or later on an Apple silicon Mac.

Open the .dmg and drag Bunyi to Applications. The .zip is the same app for anyone scripting an install.

The first generation in each mode downloads that mode’s model (about 1.5–4.5 GB) with a progress bar; after that it runs offline.

SHA-256:

c3f819edfcfde41990794d6c8a2401746b15900e60e5118c81bc4d6eb572ea3b  Bunyi-1.3.1-6.dmg
78b73bcda424fec6806594dbf8d969c2afed9a83cf599769f777315c439efb74  Bunyi-1.3.1-6.zip
0fecb482a831547a0db6d615766bc40044da7e1f83dda0c09d8ef3265ebc76ce  Bunyi-CLI-1.3.1-6-macos-arm64.zip

Bunyi for Mac 1.3.0

Bunyi 1.3.0 makes long passages dependable, brings the complete Bunyi CLI to
the Mac, and closes the gaps found in a full keyboard, VoiceOver, and feature
specification audit.

Long passages finish

• Long text is generated in bounded sections instead of exhausting the model. Bunyi keeps the selected speaker, style, or cloned voice consistent, joins the sections into one recording, and writes one History entry only after the whole passage succeeds.
• A failed section can recover. Bunyi retries with a smaller section and explains which part it is working on. Stop remains available throughout.
• Voice Design carries its designed voice into the continuation. It creates one short opening, then uses Voice Clone to preserve that voice through the rest of the passage. The first use may therefore download the separate clone model.
• Short speech finishes on the model's real end-of-speech signal. Bunyi no longer cuts off valid speech because a text-length estimate guessed too low, and it rejects genuinely unfinished output instead of saving it as though it were complete.

Voice Clone stays local and aligned

• Automatic transcription now runs locally with Whisper. Auto detects the language without sending the recording to Apple or any other service. The multilingual transcription model is about 148 MB and downloads once on first use.
• Bunyi samples the same leading, at-most-ten-second window for transcription and cloning. Longer recordings remain valid; the transcript and reference audio can no longer drift apart and leave generation running without useful speech.
• Old saved voices with a missing transcript repair themselves. Selecting one and pressing Generate recreates and saves the transcript locally. No manual transcript or language selection is required.
• Missing or mismatched clone input stops before generation. Bunyi identifies the field that needs attention instead of loading a multi-gigabyte model first.

A complete native Mac CLI

• The release now includes a separate signed and notarized Apple-silicon CLI archive. It exposes generation, transcription, speakers, saved voices, History, playback, models, configurations, Doctor, logs, and backup/restore without opening the desktop app.
• Persistent local server and detached jobs are included. Automation can keep one MLX model loaded, follow structured progress, cancel work, and recover job results through an owner-only Unix socket. It opens no network listener.
• The CLI is self-contained. Its archive carries the shared runtime, MLX Metal resources, shell completions, licence, credits, and command reference.

Downloads explain themselves

• First-use model setup shows measured bytes, speed, time left, verification, and separate file/overall progress. Waiting, stalled, retrying, and resumed transfers are named instead of looking frozen.
• Interrupted downloads resume safely. Partial files are verified before use, and a paused or unavailable Bunyi model service offers an explicit switch to Hugging Face without discarding completed data.
• Logs no longer retain credentials embedded in model-source URLs. They keep the host and useful path while redacting secrets.

Keyboard, VoiceOver, and everyday fixes

• Every window and action has a reliable keyboard route. Focus traversal, reverse traversal, History rows, playback, Settings tabs, Doctor, Logs, dialogs, and the busy-close alert were verified with real key events and VoiceOver.
• Generation progress is paced for speech. VoiceOver announces meaningful phases and sampled progress instead of being interrupted by every emitted frame.
• History remains available during generation. It can be inspected while the active generation mode stays safely locked, and Stop remains reachable there.
• Generate explains missing input in place. The button stays available, then marks and focuses the missing script, voice description, or reference clip without starting model work.
• Custom models-folder backups now have real progress and cancellation. Stop interrupts the staging copy as well as the archiver and removes only the incomplete new backup.
• About Bunyi now includes macOS, model credits, software credits, licences, and links. New WAV files also record that they were produced on macOS, while older files remain readable.

Existing downloaded speech models, saved voices, settings, and History remain in
place when updating from Bunyi 1.2.0.

---

Signed with a Developer ID certificate, notarized by Apple, and stapled.

Requires macOS 15 or later on an Apple silicon Mac.

Open the .dmg and drag Bunyi to Applications. The .zip is the same app for anyone scripting an install.

The first generation in each mode downloads that mode’s model (about 1.5–4.5 GB) with a progress bar; after that it runs offline.

SHA-256:

2f07f699f7c4d66b0ae7a988759c6b1a296bee73b33d9a8e4fd796288efc89a8  Bunyi-1.3.0-5.dmg
2a47c1024980a9599c7a61c6dddc11d47cfe002089ca9993abd039e24c916a2e  Bunyi-1.3.0-5.zip
dcba411d984e2920c54454dc22f9a373bc4faae63adc2afd9e0c1e82ae1e72e0  Bunyi-CLI-1.3.0-5-macos-arm64.zip

Bunyi for Windows and Linux 1.3.1

Features

• feat: default all packages to the Bunyi mirror (#249)

Other changes

• chore(dotnet): release 1.3.1

Full changes: dotnet-v1.3.0...dotnet-v1.3.1

---

Installing

• Windows — run Bunyi-1.3.1-win-x64-setup.exe, then open Bunyi from Start. Setup installs for the current user and offers an optional desktop shortcut.
• Windows with NVIDIA acceleration — run Bunyi-1.3.1-win-x64-cuda-setup.exe. It replaces the standard edition and keeps your data. Compatible NVIDIA drivers, CUDA and cuDNN libraries must be installed separately; Bunyi falls back to CPU if GPU acceleration cannot load. Run the standard installer to switch back.
• Ubuntu 24.04 / Debian 12 or 13 — install bunyi_1.3.1_amd64.deb using your package installer.
• Fedora 43 or 44 — install bunyi-1.3.1-1.x86_64.rpm using your package installer.

Linux packages install native dependencies and add Bunyi to the application menu. Close Bunyi before upgrading. Upgrades, repair and uninstall retain your models, voices, recordings and settings. Models download through Bunyi after installation.

The standard Windows desktop ZIP has been replaced by the setup EXE. The Linux portable tarball remains available: extract it and run ./Bunyi.App. CUDA and CLI archives remain available as portable alternatives.

Installer update — 15 September 2026: these 1.3.1 installers were built from the changes merged in #250, with Store-derived icons and metadata and Windows app-local Visual C++ runtime dependencies. The original release tag is unchanged. Build and installation checks passed on Windows and all five listed Linux distributions.

CUDA installer update — 15 September 2026: #252 adds the CUDA setup EXE and updates standard Windows setup to remove CUDA provider files when switching back. Both editions use the same Bunyi installation and saved data. Installer packaging uses commit 2c96afb. The EXEs were built and installation-tested locally; CI build and installation checks also passed for the Windows packaging changes. Both editions remain unsigned. Linux packages and the release tag are unchanged.

CLI for agents and terminals — use the separate Bunyi-CLI- archive for your OS, extract the entire folder, and run bunyi --help (Windows: bunyi.exe --help). Add that folder to PATH if desired. No desktop session or .NET installation is required. See the CLI guide for one-shot generation, the resident server, and download progress.

Self-contained: there is no .NET runtime to install.

The -cuda downloads are the same app built against ONNX Runtime with its GPU package, measured 1.7x faster on long text. Take one only if you have an NVIDIA card and the NVIDIA CUDA Toolkit installed: the provider links against cuBLAS and cudart, which are not shipped here. If CUDA cannot load, the app falls back to the CPU, says so in the log, and generates normally. Everyone else wants the plain download.

Code signing policy

Windows — not code-signed. SmartScreen warns the first time you run it: choose More info, then Run anyway. There is no certificate behind the download, so nothing would make that warning go away. Verify the download against the SHA-256 checksum below.

Linux — DEB, RPM and portable downloads are unsigned. Verify them against the SHA-256 checksums published below.

macOS is a separate download, signed with an Apple Developer ID certificate and notarized by Apple.

Team roles — committers, reviewers and approvers: Shazron Abdullah (@shazron).

This program does not transfer any information to other networked systems unless you ask it to: models are downloaded when you first use a mode, and nothing else leaves your machine.

What works

All three modes — preset voice, voice design and voice clone — plus saved voices, history, and backup and restore.

The first generation in each mode downloads the model for that mode (3.9–5.9 GB) with a progress bar and a time estimate; after that it runs offline. Doctor checks there is room before the download starts.

Requirements

A 64-bit Windows or Linux machine. Generation runs on the CPU unless you took a -cuda download and have the toolkit for it; long text can reach about 7 GB of memory on preset voice, so 8 GB is tight and 16 GB is comfortable.

SHA-256:

9eef041bc94972c6f6f06d33e43ab8bf5aafada7eff9af2ebb0b787678a32240  Bunyi-1.3.1-win-x64-cuda-setup.exe
3df269d28b3f8b077bc5164590332d5d8dca9a2d61cbe655881fd10be2370ae0  Bunyi-1.3.1-win-x64-setup.exe
ca9fb600f384dd1460beb258ce5313b03501118cae805c6a77c5ac1c2906510d  bunyi_1.3.1_amd64.deb
309cb06c2531bd5b7985040a485050500d5d3d3ea29018e6d3724c9ff621cc47  bunyi-1.3.1-1.x86_64.rpm
1258de6f715cc0b637347591b98a959d42ecc36febf1e50092aedde502719bff  Bunyi-1.3.1-linux-x64-cuda.tar.gz
b128b5481c1c605440f2f810cac4e2b83ac60a2a36bcf211be054999ac845a61  Bunyi-1.3.1-linux-x64.tar.gz
c02300b3a172345d921d59ad8d789b66b0a1179a09e0ccd69434beffcf66a00e  Bunyi-1.3.1-win-x64-cuda.zip
5d88d1f6a11ba1b1e34e4fd53cd1975f2bae0517c0fde77f99640ef6d71a195c  Bunyi-1.3.1.0-win-x64-store.msix
bf1240529baf62c28ff4990a4cd4649a6dc736cde948436e388063cf8a5abfe8  Bunyi-CLI-1.3.1-linux-x64-cuda.tar.gz
f619fea70bc12a5492f01d2b2d7fe7ffbcb64cd12f59b8fdf97f4d600867f120  Bunyi-CLI-1.3.1-linux-x64.tar.gz
256000220ca0b0d396ce2ea98eefa0d42cafe0384ed5a051f5083c220ab25674  Bunyi-CLI-1.3.1-win-x64-cuda.zip
eeb86ab2f187fe01b65fcefbeeedefa45ff9d63c7ff6fd110a144044ad84cdfc  Bunyi-CLI-1.3.1-win-x64.zip

Bunyi for Windows and Linux 1.3.0

Bunyi 1.3.0 — Windows and Linux

Longer scripts

All three modes estimate speech duration and split long scripts into smaller
sections. Attempts that run past their limit retry as shorter sections. Bunyi
saves one complete recording and then plays it. Stop cancels the unfinished
recording. Long Voice Design requests use a completed opening as a reference
for the remaining speech and may download the configured clone model.

CLI for agents and terminals

Separate CLI archives provide preset, design and clone generation, an optional
persistent server, model downloads, job progress/cancellation, local playback,
and management commands. Everything runs locally. Read the CLI guide before
using the server or automating generation.

Downloads and settings

Downloads show bytes received and distinguish waiting from slow or stalled
transfers. Rate limits use bounded retries with a countdown. If the Bunyi mirror
is unavailable, an explicit recovery action can switch the affected mode to
Hugging Face. Existing source-specific downloads are kept.

Portable desktop and CLI builds default to Hugging Face. The separately prepared
Windows MSIX defaults to the Bunyi mirror. Settings → Models can switch all
three modes between them, and the saved choice survives restart and upgrade.
Reset restores the defaults packaged with that build. Switching sources may
require separate model downloads.

The script heading and scrollbar now fit correctly in the generation view.

Availability

Desktop and CLI archives are available for Windows/Linux x64, each with a
standard CPU build and an optional NVIDIA CUDA build. The MSIX is a separate
unsigned submission/testing artifact, not a directly installable public download
or a claim of Microsoft Store certification. Native macOS releases are separate.

---

Installing

Windows — unzip and run Bunyi.App.exe.
Linux — untar and run ./Bunyi.App.

CLI for agents and terminals — use the separate Bunyi-CLI- archive for your OS, extract the entire folder, and run bunyi --help (Windows: bunyi.exe --help). Add that folder to PATH if desired. No desktop session or .NET installation is required. See the CLI guide for one-shot generation, the resident server, and download progress.

Self-contained: there is no .NET runtime to install, and nothing is written outside the folder you unpack plus your own app-data directory.

The -cuda downloads are the same app built against ONNX Runtime with its GPU package, measured 1.7x faster on long text. Take one only if you have an NVIDIA card and the NVIDIA CUDA Toolkit installed: the provider links against cuBLAS and cudart, which are not shipped here. If CUDA cannot load, the app falls back to the CPU, says so in the log, and generates normally. Everyone else wants the plain download.

Code signing policy

Windows — not code-signed. SmartScreen warns the first time you run it: choose More info, then Run anyway. There is no certificate behind the download, so nothing would make that warning go away. Verify the archive against the SHA-256 checksum below.

Linux — nothing signs a portable tarball in a way the system checks, so there is no equivalent to claim. Verify the archive against the SHA-256 checksum published below.

macOS is a separate download, signed with an Apple Developer ID certificate and notarized by Apple.

Team roles — committers, reviewers and approvers: Shazron Abdullah (@shazron).

This program does not transfer any information to other networked systems unless you ask it to: models are downloaded when you first use a mode, and nothing else leaves your machine.

What works

All three modes — preset voice, voice design and voice clone — plus saved voices, history, and backup and restore.

The first generation in each mode downloads the model for that mode (3.9–5.9 GB) with a progress bar and a time estimate; after that it runs offline. Doctor checks there is room before the download starts.

Requirements

A 64-bit Windows or Linux machine. Generation runs on the CPU unless you took a -cuda download and have the toolkit for it; long text can reach about 7 GB of memory on preset voice, so 8 GB is tight and 16 GB is comfortable.

SHA-256:

a98c9326269cdcfe4655cfdc6bd477d08e34e5da6edddadc2f799ec8c1c1eeb9  Bunyi-1.3.0-linux-x64-cuda.tar.gz
b51a6d7eb9e51b25bc55d0da7580ff32c190c2cac00553e70633cc23b93f9721  Bunyi-1.3.0-linux-x64.tar.gz
1a5386c1a55fd78abd6ef10b5840c9ea997c4444de709df320890049b5a753bf  Bunyi-1.3.0-win-x64-cuda.zip
83fd43eff800531174c5bef1952bed6a0f84b6f249779171539f42f03972e82c  Bunyi-1.3.0-win-x64.zip
24618690f5c88c1f66d87b23947b3899afb7cc2cfe22aa0cff0c5a73539cdd12  Bunyi-1.3.0.0-win-x64-store.msix
9797e4661fa120a2d4a6aadb26b89361f8243c4ea94cb1293f3e02713c5ec117  Bunyi-CLI-1.3.0-linux-x64-cuda.tar.gz
ff379a97b07dedd053819552d6bf374a49b000ccf2799becfd6badbb17b5a233  Bunyi-CLI-1.3.0-linux-x64.tar.gz
cb8a82c029648903687c0e4334957be21a6a782a4397fc91ee16cbb58c51d6b4  Bunyi-CLI-1.3.0-win-x64-cuda.zip
184907bf411c23f31a53371d348651ff5b79aa9dc04f97d3f8849cf566c0ff4e  Bunyi-CLI-1.3.0-win-x64.zip

Bunyi for Windows and Linux 1.2.0

Bunyi 1.2.0 for Windows and Linux improves keyboard access and screen-reader speech, fixes Linux audio playback, and adds an easier way to choose the Bunyi model mirror.

Accessibility and keyboard navigation

• Controls announce useful names and context. Settings actions identify the configuration, folder, or model they affect; toolbar buttons use short names while retaining their longer hover tips.
• Linux Orca now reads focused controls and voice and appearance choices, without announcing layout containers or repeating Script/Style placeholders and downloaded-model details.
• Generation announces state changes and paced frame progress. The remaining Linux Orca screen audit was manually verified before release.
• Settings keyboard navigation returns to the tab bar, and History supports arrow keys, Home, and End.
• Open Settings with Ctrl+,, Logs with Ctrl+L, Doctor with Ctrl+D, and Help with F1.

Linux audio

• Fix audio backend selection that could prevent playback with an InvalidArgs device error, and report the actual native backend correctly in Logs.

Model sources

• Settings -> Models includes a Bunyi mirror configuration. Select Restore to use it for all three modes; Hugging Face remains the default.
• Saved configurations show where their sources point, and the Restore action is clearly labelled.
• Model-source fields show their actual defaults, and source changes take effect without restarting the app.
• For mirror operators, the source repository includes a tool to create manifests from an existing downloaded-model folder.

Includes the fixes from PR #205 and PR #206.

---

Installing

Windows — unzip and run Bunyi.App.exe.
Linux — untar and run ./Bunyi.App.

Self-contained: there is no .NET runtime to install, and nothing is written outside the folder you unpack plus your own app-data directory.

The -cuda downloads are the same app built against ONNX Runtime with its GPU package, measured 1.7x faster on long text. Take one only if you have an NVIDIA card and the NVIDIA CUDA Toolkit installed: the provider links against cuBLAS and cudart, which are not shipped here. If CUDA cannot load, the app falls back to the CPU, says so in the log, and generates normally. Everyone else wants the plain download.

Code signing policy

Windows — not code-signed. SmartScreen warns the first time you run it: choose More info, then Run anyway. There is no certificate behind the download, so nothing would make that warning go away. Verify the archive against the SHA-256 checksum below.

Linux — nothing signs a portable tarball in a way the system checks, so there is no equivalent to claim. Verify the archive against the SHA-256 checksum published below.

macOS is a separate download, signed with an Apple Developer ID certificate and notarized by Apple.

Team roles — committers, reviewers and approvers: Shazron Abdullah (@shazron).

This program does not transfer any information to other networked systems unless you ask it to: models are downloaded when you first use a mode, and nothing else leaves your machine.

What works

All three modes — preset voice, voice design and voice clone — plus saved voices, history, and backup and restore.

The first generation in each mode downloads the model for that mode (3.9–5.9 GB) with a progress bar and a time estimate; after that it runs offline. Doctor checks there is room before the download starts.

Requirements

A 64-bit Windows or Linux machine. Generation runs on the CPU unless you took a -cuda download and have the toolkit for it; long text can reach about 7 GB of memory on preset voice, so 8 GB is tight and 16 GB is comfortable.

SHA-256:

6edbd41127b90dcea445e8f0a0b0619d4fe00c771ea05857bef08ce3c0babd7c  Bunyi-1.2.0-linux-x64-cuda.tar.gz
c9175a08ee000da9c024aab3982edfcb0195414141dbd0c62060c5f8ba3c31e9  Bunyi-1.2.0-linux-x64.tar.gz
9821fd210bc89aa5d1bbbee0b39f421d7a3565da70cdafd2e64e31e5b4cccb25  Bunyi-1.2.0-win-x64-cuda.zip
edd298492c2aaabfd6c0f2e1c5aab0558e832279aa7dbd4973b4234beed6f1f3  Bunyi-1.2.0-win-x64.zip