feat: 集成Piper自建TTS并增加朗读回退

server/app.py

@@ -11,6 +11,7 @@ from server.auth import admin_required, authenticate, create_user, login_require
 from server.db import BASE_DIR, get_db, init_db
 from server.entries_api import entries_bp
 from server.settings_api import settings_bp
+from server.tts_api import tts_bp
 from server.vocab_api import vocab_bp
 
 
@@ -28,6 +29,7 @@ def create_app():
     app.register_blueprint(entries_bp)
     app.register_blueprint(vocab_bp)
     app.register_blueprint(settings_bp)
+    app.register_blueprint(tts_bp)
 
     @app.post("/api/register")
     def register():

server/tts_api.py

@@ -0,0 +1,40 @@
+from flask import Blueprint, jsonify, request, send_file
+
+from server.auth import login_required
+from server.tts_service import generate_speech, piper_available
+
+tts_bp = Blueprint("tts", __name__)
+
+
+@tts_bp.route("/api/tts", methods=["GET", "POST"])
+@login_required
+def tts_audio(user):
+    if request.method == "POST":
+        data = request.get_json(silent=True) or {}
+        text = (data.get("text") or "").strip()
+        lang = (data.get("lang") or "es").strip().lower()
+    else:
+        text = (request.args.get("text") or "").strip()
+        lang = (request.args.get("lang") or "es").strip().lower()
+
+    if not text:
+        return jsonify({"error": "朗读内容不能为空"}), 400
+    if len(text) > 1000:
+        return jsonify({"error": "朗读内容不能超过 1000 个字符"}), 400
+    if lang not in ("es", "en", "zh"):
+        return jsonify({"error": "不支持的朗读语言"}), 400
+    if not piper_available():
+        return jsonify({"error": "Piper 未配置或未安装"}), 503
+
+    try:
+        audio_path = generate_speech(text, lang)
+    except Exception as exc:
+        return jsonify({"error": "Piper 生成失败: " + str(exc)}), 500
+
+    return send_file(
+        audio_path,
+        mimetype="audio/wav",
+        as_attachment=False,
+        download_name="piper.wav",
+        conditional=True,
+    )

server/tts_service.py

@@ -0,0 +1,59 @@
+import hashlib
+import os
+import subprocess
+
+SERVER_DIR = os.path.dirname(os.path.abspath(__file__))
+DEFAULT_CACHE_DIR = os.path.join(SERVER_DIR, "data", "tts_cache")
+
+PIPER_BIN = os.environ.get("PIPER_BIN", "/opt/piper/venv/bin/piper")
+PIPER_MODELS_DIR = os.environ.get("PIPER_MODELS_DIR", "/opt/piper/voices")
+TTS_CACHE_DIR = os.environ.get("TTS_CACHE_DIR", DEFAULT_CACHE_DIR)
+
+VOICES = {
+    "es": "es_ES-davefx-medium.onnx",
+    "en": "en_US-lessac-medium.onnx",
+    "zh": "zh_CN-huayan-medium.onnx",
+}
+
+
+def piper_available():
+    return bool(PIPER_BIN) and os.path.isfile(PIPER_BIN)
+
+
+def generate_speech(text, lang="es"):
+    voice = VOICES.get(lang, VOICES["en"])
+    digest = hashlib.sha256((lang + "\0" + text).encode("utf-8")).hexdigest()[:24]
+    cache_path = os.path.join(TTS_CACHE_DIR, f"{lang}_{digest}.wav")
+    if os.path.isfile(cache_path):
+        return cache_path
+
+    model_path = os.path.join(PIPER_MODELS_DIR, voice)
+    if not os.path.isfile(model_path):
+        raise RuntimeError(
+            f"Piper 模型不存在: {model_path}，请先下载语音模型并配置 PIPER_MODELS_DIR"
+        )
+
+    os.makedirs(TTS_CACHE_DIR, exist_ok=True)
+    tmp_path = cache_path + ".tmp"
+    try:
+        proc = subprocess.run(
+            [
+                PIPER_BIN,
+                "--model",
+                model_path,
+                "--output_file",
+                tmp_path,
+            ],
+            input=text.encode("utf-8"),
+            capture_output=True,
+            timeout=90,
+        )
+    except subprocess.TimeoutExpired as exc:
+        raise RuntimeError("Piper 生成超时") from exc
+
+    if proc.returncode != 0 or not os.path.isfile(tmp_path):
+        stderr = proc.stderr.decode("utf-8", errors="replace")
+        raise RuntimeError(stderr or "Piper 生成失败")
+
+    os.replace(tmp_path, cache_path)
+    return cache_path