Repository navigation
Releases
Releases · nitpacker/EarPurifier
EarPurifier v1.0.0 - First Stable Release
v0.0.3 - Checking Verification Status
هدوء الإصدار v0.03
تحديث جديد لتحسين تجربة المستخدم.
✅ التحقق من حالة توثيق الحساب برقم الهاتف، وإشعار المستخدم في حالة استخدام نموذج يتطلب الوصول إلى كروت المعالجة عند استخدام كاجل.
🎨 إصلاح خلل في عرض زر تحميل النماذج محلّيًا.
EarPurifier v0.0.3
New update focusing on improving the user experience.
✅ Warning the user when a GPU-requiring model is selected if the account is not verified with a phone number.
🎨 Fixing a display issue with the "Download Model" button.
v0.0.2 - Local Mode Improvements.
هدوء الإصدار v0.0.2
تحديث جديد يركز على تسريع وتحسين المحرك المحلي، وإعادة تصميم دقة حساب الوقت المتبقي (ETA)، وتحسينات لواجهة المستخدم والترجمة.
⚡ المحرك المحلي والتثبيت
- تثبيت أسرع بحجم أصغر: تقليص حزم المحرك المحلي بحذف التبعيات غير الضرورية وإضافة تفريغ الذاكرة المؤقتة، مع عرض دقيق للحجم (~350 ميجابايت للمعالج و ~2.7 جيجابايت لكروت شاشة NVIDIA).
- إصلاح التراجع التلقائي (CUDA Fallback): تحسين التثبيت ليتراجع بسلاسة إلى نسخة المعالج في حال تعذر تثبيت تسريع كرت الشاشة.
- التحقق الذكي من التثبيت: إضافة مؤشر حالة ذري لمنع إعادة تنزيل المحرك مع دعم التعرف التلقائي على التثبيتات السابقة.
- تسريع معالجة الملفات: استخراج مسار الصوت البشري فقط وحذف الخطوات الزائدة لتوفير مساحة التخزين وزمن المعالجة.
⏱️ دقة شريط التقدم والوقت المتبقي (ETA)
- محرك وقت متبقي ذكي ومرن: حساب الوقت بناءً على مواصفات العتاد وطول الملف الصوتي، ثم التحديث المباشر أثناء المعالجة بناءً على سرعة الإنجاز الفعلية.
- شريط تقدم متصل (0–100%): منع تراجع شريط التقدم إلى الخلف، ودعم دقيق للمرحلتين في نموذج Kim Vocal 2.
- حماية حالة المهام: ضمان عدم استبدال حالة المهمة (مكتملة / ملغاة) بأي رسائل متأخرة من العمليات الخلفية.
🎨 واجهة المستخدم وتجربة الاستخدام
- مؤشر تنزيل متطور: إضافة مؤشر دائري متحرك ومراحل واضحة أثناء تثبيت المحرك المحلي (المرحلة 1 من 3، إلخ).
- شريط تقدم متوافق مع الاتجاهات (RTL / LTR): عرض التدرج اللوني لأشرطة التقدم بشكل متوافق تمامًا مع الواجهة العربية والإنجليزية.
- تسميات واضحة ومبسطة: تحسين نصوص الواجهة والرسائل الإرشادية وتوضيح رسالة أوقات الذروة للخوادم السحابية.
EarPurifier v0.0.2
This release brings major lifecycle optimizations to the local processing engine, a redesigned two-phase ETA estimation engine, seamless RTL/LTR progress bar visuals, and localized copy refinements.
⚡ Local Engine & Dependency Lifecycle
- Faster Setup & Lighter Footprint: Pruned redundant dependencies (
torchvision/torchaudio) and introduced--no-cache-dir. Engine download sizes are now accurately tailored to detected hardware (~350 MB on CPU vs ~2.7 GB on CUDA). - Robust CUDA Fallback: Subprocess handling now seamlessly degrades to CPU PyTorch if CUDA installation fails.
- Atomic Environment Verification: Added
.engine_installedmarker file with backward-compatible auto-detection to prevent redundant engine re-downloads. - Optimized Separation: Enabled single-stem export (
--single_stem=Vocals) to eliminate unnecessary instrumental file writes.
⏱️ Progress Tracking & Dynamic ETA
- Two-Phase Hardware-Aware ETA: Computes initial estimates based on audio length and hardware benchmarks, then switches to live throughput rate sampling.
- Monotonic 0–100% Progress: Fixed progress bar regression across pipeline stages, including full two-pass tracking for Kim Vocal 2.
- Terminal State Guard: Terminal job card states (Completed, Error, Cancelled) are strictly protected against trailing log output.
🎨 UI & Localization Refinements
- Animated Milestone Overlay: Integrated
RotatingSpinnerand 3-step milestones during engine initialization. - Direction-Aware Progress: Progress bars and model download progress indicators now natively support both RTL (Arabic) and LTR (English) layouts.
- Refined Copy: Updated window titles, streamlined dialog explanations, and improved peak-hour cloud queue messaging.