LocalDub v0.3.3 preview

LocalDub 0.3.3 preview for Apple Silicon, macOS 26 or newer.

Developer ID signed, Apple notarized and stapled. Gatekeeper accepts the app
extracted from the final ZIP with quarantine enabled; no security bypass is
required. Includes optional local voice cleanup, engine-driven language
selection, audio file and URL inputs, and sequential multilingual queues.
Source plugins are installed separately. Inference runs locally.

The exact published archive now passes automatic setup from an empty workspace:
Python, dependencies and all three default models downloaded and verified.
No existing runtime or model cache was copied. Cold setup took about 31 minutes
on the test connection, followed by four complete short Mandarin/Spanish dubs,
six speech checks and eight full MP3/WAV decodes. Batch pause, full restart,
resume, export checksums and invalid-input rejection pass. Previous file-input
and cancellation/resume checks also pass. See VALIDATION.md and notarization.json
for the complete verification scope and limitations.

Bilingual listening and emotional-fidelity review remain pending. Representative
resource usage and full-catalogue performance remain unaccepted. Automated speech
checks do not establish translation quality or preservation of the original voice.
Fresh workspace testing used the existing M4 Mac, not a wiped operating system.

Extract the ZIP, copy LocalDub.app to Applications and open it. Allow at least
25 GB free; setup downloads approximately 8 GB of models plus Python and packages.
Keep the selected storage attached and the Mac awake during processing.

Verified update - 23 October 2026

The existing signed and notarized ZIP is already the latest verified app build;
its download URL and SHA-256 remain unchanged. Additional sequential
Mandarin/Spanish, audio/transcript integrity and in-place upgrade checks pass.
Existing models, settings, jobs and generated audio were preserved during the
installed-app upgrade. Read INSTALL_UPGRADE.md and the updated VALIDATION.md.
Private plugins and their account configuration remain separately distributed
and are not included in this public GitHub release.

LocalDub v0.3.2 preview

Fixes first-run model-download progress for current Hugging Face partial filenames and clarifies retry behavior. Completed model files are reused; interrupted files may restart.

Validated with a fresh workspace: automatic runtime and model setup, offline private integration installation, two complete real podcasts dubbed sequentially to Mandarin and Spanish (four new outputs), six speech checks, eight complete audio decodes, paused restart/resume and idempotent exports. 148 public tests and15 private adapter tests pass. Generic file/cancellation acceptance carries forward from0.3.0; inference paths are unchanged.

Preview only: ad-hoc signed and not notarized. Gatekeeper rejects this build for standard downloaded-app approval; no security bypass was used. Developer ID signing/notarization, independent bilingual listening approval and matched resource benchmarking remain pending. Private integrations, credentials and client recordings are excluded.

LocalDub v0.3.1 preview

Adds an optional local export extension for separately installed private integrations. The generic audio bundle is retained if an extension fails, and exports can be retried after configuration repair. Successful retries update the saved export result.

Validation: 148 public tests, 15 private adapter tests, four real existing audio exports and native app export requests; all 91 existing export payload hashes remained unchanged. This packaging update does not change inference; input, batch and cancellation/resume acceptance carries forward from 0.3.0.

Preview: ad-hoc signed, not notarized. Independent client first launch, expressive bilingual listening approval and matched resource benchmarking remain pending. Private plugins, credentials and client data are excluded.

LocalDub v0.3.0 preview

Import an entire podcast collection through a separately installed private catalogue plugin, select an engine-supported output language and run one episode at a time. Metadata imports completely before the queue is created; each audio downloads, dubs and exports before the next starts. The public app contains no private adapters or credentials.

Large queues now use paged job cards and cached status summaries. Durable batches support pause/retry, reject partial catalogue imports and pause before another episode when storage falls below 10 GiB. Local exports retain title/language/version folders, immutable checksums and collision-resistant names.

Validated with 146 automated tests, a 2,000-item queue regression, complete live metadata imports, four real catalogue-generated Mandarin/Spanish outputs and a file-upload cancel/resume run. Export checksums and full MP3/WAV decoding passed. Existing completed jobs were preserved through the native upgrade.

Preview limitations: ad-hoc signed, not notarized; browser-quarantined first launch remains unverified. Full-collection inference, human bilingual/emotion acceptance and matched resource benchmarks remain pending. Keep the app open, the Mac awake and storage attached. Private plugins are distributed and configured separately. No cloud uploads or podcast publishing are enabled.

LocalDub v0.2.0 preview

Language selection now follows the selected speech and translation engines. Qwen3-TTS exposes ten supported languages; the optional IndexTTS 2.5 adapter exposes five when separately configured. URL batches can select additional supported languages. Unsupported combinations fail before queuing, and matching source/target languages retain the source transcript.

The Mac bundle includes a reproducible icon at standard and Retina sizes, Finder/Dock metadata and a native About identity. The setup screen no longer limits its description to Mandarin and Spanish.

Validation: 140 tests on Mac and GitHub CI, 15 live API checks, native French file dubbing and English URL dubbing with full output decoding and successful speech checks. Three-language URL queue controls were tested without rendering those queue-test jobs. Ten existing jobs and 91 exported payloads remain intact. See validation.json for exact scope.

This remains a preview: ad-hoc signed and not notarized. Browser-quarantined first launch, human listening acceptance across languages and matched resource benchmarks remain pending. Automated speech checks do not prove translation meaning, original emotion or identical voice. Private integrations and credentials are excluded from the public source and installer.

LocalDub v0.1.1 preview

LocalDub 0.1.1 improves setup and download recovery on Apple Silicon.

• Restore required packages when a moved or recreated workspace still has an earlier setup receipt.
• Count only the selected model files during resumed download progress.
• Try alternate verified public addresses when a connection fails, preserving TLS hostname and private-network checks.
• Hide the busy indicator after the engine stops and keep installer version metadata consistent.

Validation: 133 local tests and 133 hosted CI tests passed. A fresh-install QA run completed a file upload plus a four-output Mandarin/Spanish URL batch using short public speech fixtures. All five speech checks passed, all MP3s decoded, and 35 exported payload hashes matched. Cached native restart was verified with external IP connections denied. See validation.json for exact scope, timing and recovery checks.

This remains a preview: the app is ad-hoc signed and not notarized. Normal quarantined browser installation, human bilingual/emotion listening acceptance and broader resource benchmarks remain pending. Automated speech readback does not establish meaning, voice identity or emotional fidelity.

LocalDub v0.1.0 preview

LocalDub is a local audio-dubbing app for Apple Silicon Macs. This first public preview supports English audio to Mandarin Chinese and Spanish using on-device speech recognition, translation and reference-voice synthesis.

Download and install

1. Download LocalDub-0.1.0-arm64.zip and verify it against SHA256SUMS.
2. Extract it and copy LocalDub.app to Applications.
3. Open the app, choose a workspace, and allow first setup to download Python, packages and approximately 7.84 GB of models. Allow at least 25 GB free.

Requires Apple Silicon and macOS 26+. This preview is ad-hoc signed and not notarized; macOS may block opening it. Do not disable Gatekeeper. The repository also documents running from source.

Included

• Local audio-file and direct HTTP(S) URL inputs, optional public Podbean episode resolution, and separately installed source plugins.
• Sequential batches with up to 200 URLs or 20 uploaded files; multilingual URL queues generate both supported languages.
• Voice, pacing, pauses and audio configuration; cancellation/resume, review corrections and selective speech regeneration.
• Versioned local MP3/WAV audio, WebVTT captions and review reports, with organized podcast/language folders for URL batches.
• Automatic private runtime/package/model setup with native progress controls. Inference runs locally.

Verification

128 core tests passed locally and in hosted Linux CI. The release app launched on a 24 GB M4 Mac mini and restored completed work. Two complete recordings in both languages produced four outputs with 298/298 automated speech checks after a targeted numeric repair; all 28 final export files matched their hashes. Bundled backend resources match that inference-tested build. The source history scan found no credentials. Private source adapters, account configuration, recordings and models are excluded from this release.

See candidate.json, acceptance.json and validation.json for the exact commit, archive checksum, verification scope and pending checks. Media-tool licenses and corresponding sources are included inside the app; model licenses remain separate.

Preview limits

Human bilingual listening acceptance for meaning, emotion, pauses and voice likeness is pending. Clean setup and disconnected-network recovery on this exact ZIP, and controlled resource measurements of this candidate, are also pending. Automated readback does not establish emotional fidelity or translation quality. Lower-memory Macs and other operating systems are not certified. IndexTTS is experimental and is not installed by default. Cloud uploads and private podcast-account management are deferred.

README and setup · Validation · Issues