synth_batch.py

#!/usr/bin/env python3
import argparse, os, sys, re
import numpy as np
import soundfile as sf
from tqdm import tqdm
from utils.text import normalize_text, split_into_paragraphs, maybe_ruaccent
from utils.audio import crossfade_concat, normalize_lufs, save_wav
from backends.espeech_backend import EspeechBackend

def read_input(path: str) -> str:
    if path.lower().endswith(".txt"):
        with open(path, "r", encoding="utf-8") as f:
            return f.read()
    elif path.lower().endswith(".epub"):
        try:
            from ebooklib import epub
            from bs4 import BeautifulSoup
        except Exception:
            print("Для EPUB установите: pip install ebooklib beautifulsoup4 lxml", file=sys.stderr)
            sys.exit(2)
        book = epub.read_epub(path)
        texts = []
        for item in book.get_items():
            if item.get_type() == 9:  # DOCUMENT
                soup = BeautifulSoup(item.get_body_content(), "lxml")
                texts.append(soup.get_text(" ", strip=True))
        return "\n\n".join(texts)
    else:
        raise ValueError("Поддерживаются .txt и .epub")

def load_default_ref_text():
    """Load default reference text from local sample"""
    try:
        with open("samples/001/sample.text", "r", encoding="utf-8") as f:
            return f.read().strip()
    except FileNotFoundError:
        return ""

def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--input", required=True, help="Путь к TXT/EPUB")
    ap.add_argument("--outdir", required=True, help="Каталог для результата")
    ap.add_argument("--ref-audio", required=False, default="samples/001/sample.mp3", help="Путь к референс-аудио (6–12с)")
    ap.add_argument("--ref-text", required=False, default=load_default_ref_text(), help="Референс-текст")
    ap.add_argument("--model-repo", default=os.getenv("MODEL_REPO", "ESpeech/ESpeech-TTS-1_RL-V2"))
    ap.add_argument("--speed", type=float, default=1.0)
    ap.add_argument("--nfe-steps", type=int, default=48)
    ap.add_argument("--crossfade-ms", type=int, default=150)
    ap.add_argument("--target-lufs", type=float, default=-20.0)
    args = ap.parse_args()

    os.makedirs(args.outdir, exist_ok=True)

    backend = EspeechBackend(model_id=args.model_repo)

    raw = read_input(args.input)
    text = normalize_text(raw)
    paragraphs = split_into_paragraphs(text)
    paragraphs = [maybe_ruaccent(p) for p in paragraphs]

    print(f"Абзацев: {len(paragraphs)}")

    pieces = []
    sr = None
    for i, para in enumerate(tqdm(paragraphs, desc="Генерация")):
        audio, sr = backend.synthesize(
            text=para,
            ref_audio_path=args.ref_audio,
            ref_text=args.ref_text,
            speed=args.speed,
            nfe_steps=args.nfe_steps,
            seed=None,
        )
        pieces.append(audio)
        # Сохраняем черновики по абзацам (опционально)
        # sf.write(os.path.join(args.outdir, f"para_{i:05d}.wav"), audio, sr)

    final = crossfade_concat(pieces, crossfade_ms=args.crossfade_ms, sample_rate=sr)
    final = normalize_lufs(final, sr, target_lufs=args.target_lufs)

    out_path = os.path.join(args.outdir, "book.wav")
    save_wav(out_path, final, sr)
    print(f"ГОТОВО: {out_path}")

if __name__ == "__main__":
    main()