Step 1 · Stem separation (GPU)
idle
Step 2 · Transcription (Whisper + WhisperX)
idle
♪ Melody · pitch (RMVPE, from the original)
idle
Step 3 · Translation (DeepSeek + D-score)
idle
Step 4 · Phoneme alignment (espeak-ng IPA)
idle
Step 5 · Lyric gate (D1–D5 verdict)
idle
Step 6 · Synthesis (TCSinger, en/zh)
idle
Step 7 · Merge (final master)
idle