Skip to content

WhisperSubTranslate v2.4.4

Choose a tag to compare

@github-actions github-actions released this 08 Aug 13:44
· 23 commits to main since this release

Select your language: Click the toggle (▼) below to expand your preferred language.
언어 선택: 아래 토글(▼)을 클릭해 원하는 언어를 펼쳐 보세요.
言語選択: 下のトグル(▼)をクリックして希望の言語を展開してください。
选择语言: 点击下方折叠按钮(▼)展开您的首选语言。
Wybór języka: Kliknij przełącznik (▼) poniżej, aby rozwinąć preferowany język.

한국어 (KO)

개요

WhisperSubTranslate v2.4.4 — 안정성 패치 릴리즈. 번역·다운로드·설정·파일 보존 경계를 강화하고 Windows 실제 음성 추출까지 검증했습니다.

다운로드

  • Windows 포터블(권장): WhisperSubTranslate-v2.4.4-win-x64.zip
    → 압축 해제 → WhisperSubTranslate.exe 실행
  • Linux/macOS: 소스에서 직접 실행 (README의 Linux Setup 참조)

🔧 수정

  • 번역 언어 전달 — 선택한 원본 언어가 직접 SRT 번역과 영상 추출 후 번역의 모든 API 경로에 전달됩니다. (#44)
  • DeepL 문맥 번역 — 앞뒤 자막 문맥을 DeepL에 전달하고, 문맥별 캐시를 충돌 없이 분리합니다. (#49)
  • DeepL·MyMemory 오류 처리 — :fx 키 힌트, rate limit 분류, 오류 문구가 자막 결과에 기록되지 않는 안전장치를 보강했습니다. (#41, #42, #43, #48)
  • 자막 병합과 VAD — 중복·1ms cue 병합과 VAD 최대 발화 길이 처리를 보강했습니다. (#46, #47)
  • Sync 모델 호환성 — GPU compute type 선택과 CUDA/CPU 폴백을 안정화했습니다. (#45)
  • 휴대용 데이터 경로 — %APPDATA% 대신 실행 파일 옆 portable-data/ 또는 환경 변수 경로를 사용할 수 있습니다. (#50)
  • 파일·설정 보존 — WAV 백업 실패 시 원본 삭제를 차단하고, 설정 저장 중 입력·비동기 로드·다운로드 취소 경합을 안전하게 처리합니다.
  • 다운로드·설치 안전성 — API key와 디스크 공간을 사전 검사하고, Sync 설치의 전체 피크 공간을 계산하며, Windows postinstall redirect와 partial 파일 정리를 보강했습니다.

🧪 검증

  • Linux 전체 smoke/E2E, i18n, lint, Prettier, npm audit, LSP 통과
  • Windows native GGML CPU 및 large-v2 Sync 실제 음성 추출 통과
  • Windows portable 패키징, CUDA runtime, ffprobe, ASAR 파일 동기화 확인
English (EN)

Overview

WhisperSubTranslate v2.4.4 — stability patch release. Strengthens translation, download, settings, and file-preservation boundaries, with real Windows speech extraction validation.

Download

  • Windows Portable (Recommended): WhisperSubTranslate-v2.4.4-win-x64.zip
    → Extract → Run WhisperSubTranslate.exe
  • Linux/macOS: Run from source (see Linux Setup in README)

🔧 Fixed

  • Source-language forwarding — the selected source language now reaches every direct-SRT and post-extraction translation API path. (#44)
  • DeepL context translation — neighbouring cues are forwarded to DeepL and context-specific cache entries are collision-resistant. (#49)
  • DeepL and MyMemory errors — improved :fx key hints, rate-limit classification, and protection against writing provider error text into subtitle output. (#41, #42, #43, #48)
  • Subtitle merging and VAD — strengthened duplicate/1 ms cue merging and maximum VAD speech-duration handling. (#46, #47)
  • Sync model compatibility — stabilized GPU compute-type selection and CUDA/CPU fallback behavior. (#45)
  • Portable data layout — use portable-data/ beside the executable or WHISPER_PORTABLE_DATA instead of %APPDATA%. (#50)
  • File and settings preservation — failed WAV backups never delete the source, while settings edits, asynchronous loads, and download cancellation races are handled safely.
  • Download and installation safety — API-key and disk-space preflights, aggregate Sync peak-space checks, Windows postinstall redirect draining, and partial-file cleanup are hardened.

🧪 Validation

  • Linux smoke/E2E, i18n, lint, Prettier, npm audit, and LSP checks passed
  • Real Windows GGML CPU and large-v2 Sync speech extraction passed
  • Windows portable packaging, CUDA runtime, ffprobe, and ASAR synchronization verified
日本語 (JA)

概要

WhisperSubTranslate v2.4.4 — 安定性パッチリリース。翻訳、ダウンロード、設定、ファイル保護を強化し、Windowsで実際の音声抽出まで検証しました。

ダウンロード

  • Windowsポータブル(推奨): WhisperSubTranslate-v2.4.4-win-x64.zip
    → 解凍 → WhisperSubTranslate.exe を実行
  • Linux/macOS: ソースから実行(READMEのLinux Setupを参照)

🔧 修正

  • 選択した原言語がすべての翻訳API経路に渡るよう修正しました。(#44)
  • DeepLに前後の字幕文脈を渡し、文脈ごとのキャッシュを安全に分離しました。(#49)
  • DeepL/MyMemoryのキー、レート制限、エラー出力の扱いを改善しました。(#41, #42, #43, #48)
  • 字幕マージ、1ms cue、VADの最大発話時間を強化しました。(#46, #47)
  • SyncモデルのGPU compute typeとCUDA/CPUフォールバックを安定化しました。(#45)
  • portable-data/ と WHISPER_PORTABLE_DATA によるポータブルデータ配置を追加しました。(#50)
  • WAVバックアップ失敗、設定入力、非同期ロード、ダウンロード中断時のデータ保護を強化しました。

🧪 検証

  • Linux smoke/E2E、i18n、lint、Prettier、npm audit、LSPを通過
  • Windows native GGML CPUとlarge-v2 Syncの実音声抽出を通過
  • Windowsポータブルパッケージ、CUDA runtime、ffprobe、ASAR同期を確認
中文 (ZH)

概述

WhisperSubTranslate v2.4.4 — 稳定性补丁版本。加强翻译、下载、设置和文件保护,并完成 Windows 实际语音提取验证。

下载

  • Windows 便携版(推荐):WhisperSubTranslate-v2.4.4-win-x64.zip
    → 解压 → 运行 WhisperSubTranslate.exe
  • Linux/macOS:从源码运行(参见 README 中的 Linux Setup)

🔧 修复

  • 修复原语言无法传递到所有翻译 API 路径的问题。(#44)
  • DeepL 现在使用前后字幕上下文,并安全隔离不同上下文的缓存。(#49)
  • 改进 DeepL/MyMemory 的密钥提示、限流识别和错误文本保护。(#41, #42, #43, #48)
  • 加强字幕合并、1ms cue 和 VAD 最大语音时长处理。(#46, #47)
  • 稳定 Sync 模型的 GPU compute type 选择和 CUDA/CPU 回退。(#45)
  • 支持 portable-data/ 和 WHISPER_PORTABLE_DATA 便携数据目录。(#50)
  • 加强 WAV 备份失败、设置编辑、异步加载和下载取消时的数据保护。

🧪 验证

  • Linux smoke/E2E、i18n、lint、Prettier、npm audit、LSP 通过
  • Windows native GGML CPU 和 large-v2 Sync 实际语音提取通过
  • Windows 便携打包、CUDA runtime、ffprobe 和 ASAR 同步已确认
Polski (PL)

Przegląd

WhisperSubTranslate v2.4.4 — wydanie poprawkowe stabilności. Wzmocniono obsługę tłumaczeń, pobierania, ustawień i ochrony plików oraz zweryfikowano rzeczywiste rozpoznawanie mowy w Windows.

Pobieranie

  • Windows Portable (zalecane): WhisperSubTranslate-v2.4.4-win-x64.zip
    → Rozpakuj → Uruchom WhisperSubTranslate.exe
  • Linux/macOS: Uruchom ze źródeł (zobacz Linux Setup w README)

🔧 Poprawki

  • Wybrany język źródłowy jest teraz przekazywany do wszystkich ścieżek API tłumaczenia. (#44)
  • DeepL otrzymuje kontekst sąsiednich napisów, a pamięć podręczna jest bezpiecznie rozdzielona. (#49)
  • Ulepszono podpowiedzi kluczy, wykrywanie limitów i ochronę przed zapisem błędów MyMemory/DeepL do napisów. (#41, #42, #43, #48)
  • Wzmocniono scalanie napisów, obsługę cue 1ms i maksymalny czas mowy VAD. (#46, #47)
  • Ustabilizowano wybór compute type oraz powrót CUDA/CPU dla Sync. (#45)
  • Dodano układ danych portable-data/ i WHISPER_PORTABLE_DATA. (#50)
  • Wzmocniono ochronę danych podczas błędów kopii WAV, edycji ustawień, opóźnionego ładowania i anulowania pobierania.

🧪 Weryfikacja

  • Smoke/E2E, i18n, lint, Prettier, npm audit i LSP przeszły pomyślnie
  • Rzeczywiste rozpoznawanie mowy Windows dla GGML CPU i large-v2 Sync przeszło pomyślnie
  • Zweryfikowano pakiet Windows Portable, runtime CUDA, ffprobe i synchronizację ASAR