clean English UI with professional layout

optimize: parallel ffmpeg (8 workers), larger ASR batches (64), concurrent download fragments

restore Qwen3-ASR with Zero GPU, 3-column dataset output

use /data for persistent storage if available

fix: disable tqdm tracking to prevent float progress validation error

switch to faster-whisper CPU, remove all GPU dependencies

add error reporting for ASR phase, validate segments before transcription

replace silero VAD with energy-based VAD, zero CUDA dependency for segmentation

use persistent home dir for sessions, check disk for existing downloads

use silero-vad pip package with ONNX for CPU-only VAD, no CUDA needed

fix: wrap VAD in spaces.GPU to resolve torchaudio CUDA dependency

add checkpoint/resume system, upgrade to gradio 5.18, auto-cleanup

update gradio sdk to 5.18.0, fix launch config and jinja2 compatibility

fix: resolve jinja2 template cache error and gradio launch config

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload packages.txt with huggingface_hub

Upload folder using huggingface_hub

initial commit