60 lines
1.8 KiB
Python
60 lines
1.8 KiB
Python
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
|