#!/usr/bin/env python3 # -*- coding: utf-8 -*- """ CVNSS4.0 Vietnamese TTS Studio - Fix: Python 3.12 Compatibility (Missing 'imp' module) - Fix: Auto-download 'src' """ import os import sys import types import importlib import importlib.util import hashlib import tempfile import json import time import glob import re import shutil import subprocess from pathlib import Path # ========================================================= # 0) CRITICAL PATCH: FIX PYTHON 3.12 'imp' ERROR # ========================================================= # Python 3.12 đã xóa module 'imp', nhưng code TTS cũ vẫn gọi nó. # Đoạn code này tạo ra một module 'imp' giả để đánh lừa hệ thống. try: import imp except ImportError: print("🔧 Đang vá lỗi tương thích Python 3.12+ (Mocking 'imp' module)...") imp = types.ModuleType('imp') imp.new_module = types.ModuleType imp.reload = importlib.reload # Hàm find_module giả lập (thường được dùng trong code cũ) def _find_module(name, path=None): spec = importlib.util.find_spec(name, path) if spec is None: raise ImportError(f"No module named {name}") return (None, spec.origin, ("", "", 0)) # Trả về tuple giả imp.find_module = _find_module # Hàm load_module giả lập def _load_module(name, file, pathname, description): spec = importlib.util.spec_from_file_location(name, pathname) module = importlib.util.module_from_spec(spec) spec.loader.exec_module(module) sys.modules[name] = module return module imp.load_module = _load_module sys.modules['imp'] = imp print("✅ Đã vá xong module 'imp'.") # Các import nặng khác sau khi đã vá lỗi import torch import numpy as np import soundfile as sf import gradio as gr # ========================================================= # 1) AUTO-HEALING: DOWNLOAD MISSING CORE MODULES # ========================================================= def setup_environment(): """Tự động tải thư mục src nếu bị thiếu""" if not os.path.exists("src"): print("🔄 Phát hiện thiếu thư mục 'src'. Đang tải mã nguồn cốt lõi...") try: # Clone repo chứa src subprocess.run( ["git", "clone", "https://huggingface.co/spaces/valtecAI-team/valtec-vietnamese-tts", "temp_repo"], check=True ) # Di chuyển thư mục src if os.path.exists("temp_repo/src"): if os.path.exists("src"): shutil.rmtree("src") # Xóa cũ nếu lỗi shutil.move("temp_repo/src", "./src") print("✅ Đã cài đặt xong 'src'.") else: print("❌ Không tìm thấy 'src' trong repo đã tải.") shutil.rmtree("temp_repo", ignore_errors=True) except Exception as e: print(f"❌ Lỗi khi tải mã nguồn: {e}") # Chạy setup setup_environment() # Add src to path sys.path.insert(0, str(Path(__file__).parent)) # Import core modules try: # Import theo thứ tự để tránh lỗi vòng lặp from src.text.symbols import symbols from src.vietnamese.text_processor import process_vietnamese_text from src.vietnamese.phonemizer import text_to_phonemes, VIPHONEME_AVAILABLE from src.models.synthesizer import SynthesizerTrn print("✅ Core modules imported successfully.") except ImportError as e: print(f"🔥 Vẫn còn lỗi Import: {e}") # Fallback objects VIPHONEME_AVAILABLE = False symbols = [] SynthesizerTrn = None # ========================================================= # 2) SYSTEM CONFIGURATION & CSS # ========================================================= NEON_CSS = r""" :root { --bg-dark: #0f172a; --text-primary: #e2e8f0; --neon-cyan: #06b6d4; } body, .gradio-container, .app { background: radial-gradient(circle at 50% 0%, #1e293b 0%, #0f172a 100%) !important; color: white !important; } .panelNeon { background: rgba(30, 41, 59, 0.7); border: 1px solid rgba(255,255,255,0.08); border-radius: 16px; padding: 20px; margin-bottom: 20px; } .panelNeon textarea, .panelNeon input, .panelNeon select { background: #f1f5f9 !important; color: #0f4c81 !important; font-weight: 500; } button.primary { background: linear-gradient(135deg, #06b6d4 0%, #3b82f6 100%) !important; color: white; font-weight: bold; border: none; } """ # ========================================================= # 3) UTILITIES # ========================================================= def md5_key(*parts): return hashlib.md5("|".join(parts).encode("utf-8")).hexdigest() def find_latest_checkpoint(model_dir, prefix="G"): pattern = os.path.join(model_dir, f"{prefix}*.pth") checkpoints = glob.glob(pattern) if not checkpoints: return None checkpoints.sort(key=lambda x: int(re.search(rf"{prefix}(\d+)\.pth", x).group(1)) if re.search(rf"{prefix}(\d+)\.pth", x) else 0, reverse=True) return checkpoints[0] def download_model(): from huggingface_hub import snapshot_download hf_repo = "valtecAI-team/valtec-tts-pretrained" cache_base = Path(os.environ.get("XDG_CACHE_HOME", Path.home() / ".cache")) if os.name == "nt": cache_base = Path(os.environ.get("LOCALAPPDATA", Path.home() / "AppData" / "Local")) model_dir = cache_base / "valtec_tts" / "models" / "vits-vietnamese" if (model_dir / "config.json").exists() and list(model_dir.glob("G_*.pth")): return str(model_dir) print(f"⬇️ Downloading Model {hf_repo}...") try: snapshot_download(repo_id=hf_repo, local_dir=str(model_dir)) return str(model_dir) except Exception as e: print(f"Lỗi tải model: {e}") return str(model_dir) # Trả về đường dẫn dù lỗi để tránh crash ngay # ========================================================= # 4) MANAGER & MODEL WRAPPER # ========================================================= class TTSManager: def __init__(self): self.device = "cuda" if torch.cuda.is_available() else "cpu" print(f"🔧 Initializing TTS on {self.device}...") if SynthesizerTrn is None: raise RuntimeError("Lỗi nghiêm trọng: Không thể tải class SynthesizerTrn do lỗi import.") self.model_dir = download_model() self.ckpt_path = find_latest_checkpoint(self.model_dir, "G") self.cfg_path = os.path.join(self.model_dir, "config.json") if not self.ckpt_path or not os.path.exists(self.cfg_path): raise FileNotFoundError("Không tìm thấy Model Checkpoint hoặc Config.") self.tts = VietnameseTTS(self.ckpt_path, self.cfg_path, self.device) self.temp_dir = Path(tempfile.gettempdir()) / "neon_tts_cache" self.temp_dir.mkdir(parents=True, exist_ok=True) def synthesize(self, text, speaker, speed): try: if not text: return None, "⚠️ Chưa nhập nội dung" key = md5_key(speaker, str(speed), text[:30]) out_path = self.temp_dir / f"{key}.wav" if out_path.exists(): return str(out_path), "✅ Lấy từ Cache" audio, sr = self.tts.synthesize(text, speaker, speed) sf.write(str(out_path), audio, sr) return str(out_path), "✅ Tạo thành công" except Exception as e: return None, f"❌ Lỗi: {str(e)}" class VietnameseTTS: def __init__(self, ckpt, cfg, device="cpu"): self.device = device with open(cfg, "r", encoding="utf-8") as f: self.config = json.load(f) self.spk2id = self.config["data"]["spk2id"] self.speakers = list(self.spk2id.keys()) self.model = SynthesizerTrn( len(symbols), self.config["data"]["filter_length"] // 2 + 1, self.config["train"]["segment_size"] // self.config["data"]["hop_length"], n_speakers=self.config["data"]["n_speakers"], **self.config["model"] ).to(self.device) state = torch.load(ckpt, map_location=self.device)["model"] self.model.load_state_dict({k.replace("module.", ""): v for k,v in state.items()}, strict=False) self.model.eval() def synthesize(self, text, speaker, speed): from src.text import cleaned_text_to_sequence from src.nn import commons norm_text = process_vietnamese_text(text) phones, tones, _ = text_to_phonemes(norm_text, use_viphoneme=VIPHONEME_AVAILABLE) phone_ids, tone_ids, lang_ids = cleaned_text_to_sequence(phones, tones, "VI") phone_ids = commons.intersperse(phone_ids, 0) tone_ids = commons.intersperse(tone_ids, 0) lang_ids = commons.intersperse(lang_ids, 0) x = torch.LongTensor(phone_ids).unsqueeze(0).to(self.device) x_len = torch.LongTensor([len(phone_ids)]).to(self.device) tone = torch.LongTensor(tone_ids).unsqueeze(0).to(self.device) lang = torch.LongTensor(lang_ids).unsqueeze(0).to(self.device) sid = torch.LongTensor([self.spk2id.get(speaker, 0)]).to(self.device) with torch.no_grad(): bert = torch.zeros(1024, len(phone_ids)).unsqueeze(0).to(self.device) ja_bert = torch.zeros(768, len(phone_ids)).unsqueeze(0).to(self.device) outputs = self.model.infer(x, x_len, sid, tone, lang, bert, ja_bert, noise_scale=0.667, noise_scale_w=0.8, length_scale=speed) audio = outputs[0][0,0].detach().cpu().numpy() return audio, self.config["data"]["sampling_rate"] # ========================================================= # 5) UI # ========================================================= def create_ui(manager: TTSManager): def run(text, spk, spd): start = time.time() path, msg = manager.synthesize(text, spk, spd) dur = time.time() - start return path, f"