requirements.txt

# ───────── Core deep‑learning stack ─────────
torch==2.1.0                      # pip install torch==2.1.0+cu118 -f https://download.pytorch.org/whl/torch_stable.html  for CUDA11.8
torchaudio==2.1.0                 # same index‑url as above for CUDA builds

# ───────── Utility / training helpers ───────
tqdm>=4.66.0
matplotlib>=3.8.0                 # used in utils/*.py for loss & latent visualizers

# ───────── NLP preprocessing ────────────────
pypinyin>=0.50.0                  # pinyin tokeniser in models/text_encoder.py

# ───────── Audio post‑processing & conversion ─
pydub>=0.25.1                     # FLAC→WAV converter (requires system ffmpeg)
openunmix
whisper
soundfile

# (Optional) experiment tracking
wandb>=0.16.0                     # uncomment if you enable logging