Fix ChatTTS recursion depth exceeded on empty generation.

Disable ensure_non_empty retries, set min_new_token, always refine text, and use per-chunk manual_seed.

Co-authored-by: Cursor <cursoragent@cursor.com>

.env.example

@@ -17,3 +17,4 @@ OLLAMA_PORT=11434
 # 8GB 显存 OOM 时可调低（合成按段切分）
 # TTS_MAX_CHARS_PER_CHUNK=150
 # TTS_MAX_NEW_TOKEN=768
+# TTS_MIN_NEW_TOKEN=16

config.py

@@ -142,6 +142,9 @@ TTS_MAX_CHARS_PER_CHUNK = _env_int("TTS_MAX_CHARS_PER_CHUNK", 200)
 # ChatTTS 单段最大生成 token（越小越省显存，长句会自动切多段）
 TTS_MAX_NEW_TOKEN = _env_int("TTS_MAX_NEW_TOKEN", 1024)
 
+# 至少生成多少 audio token 才允许结束（防止首 token EOS → 无限递归重试）
+TTS_MIN_NEW_TOKEN = _env_int("TTS_MIN_NEW_TOKEN", 16)
+
 # ---------------------------------------------------------------------------
 # 上传临时文件目录
 # ---------------------------------------------------------------------------

tts_service.py

@@ -12,6 +12,7 @@ import re
 import traceback
 import uuid
 import warnings
+from dataclasses import replace
 from datetime import datetime
 from pathlib import Path
 from typing import Any, Dict, List, Optional, Tuple
@@ -32,6 +33,7 @@ from config import (
     SPEAKER_SAMPLE_MIN_SEC,
     TTS_MAX_CHARS_PER_CHUNK,
     TTS_MAX_NEW_TOKEN,
+    TTS_MIN_NEW_TOKEN,
     TTS_SAMPLE_RATE,
     TTS_SPEED_PROMPT,
     TTS_TEMPERATURE,
@@ -641,10 +643,14 @@ def generate_voice(refined_text: str) -> Tuple[bool, str, Optional[str]]:
             top_P=TTS_TOP_P,
             top_K=TTS_TOP_K,
             max_new_token=TTS_MAX_NEW_TOKEN,
+            min_new_token=TTS_MIN_NEW_TOKEN,
+            ensure_non_empty=False,
         )
 
         params_refine_text = ChatTTS.Chat.RefineTextParams(
             prompt="[oral_2][laugh_0][break_4]",
+            ensure_non_empty=False,
+            min_new_token=4,
         )
 
         logger.info(
@@ -657,12 +663,31 @@ def generate_voice(refined_text: str) -> Tuple[bool, str, Optional[str]]:
         segment_wavs: List[np.ndarray] = []
         for idx, chunk in enumerate(chunks, start=1):
             release_cuda_cache()
-            wavs = chat.infer(
-                chunk,
-                skip_refine_text=(idx > 1),
-                params_refine_text=params_refine_text,
-                params_infer_code=params_infer_code,
-            )
+            # manual_seed 每段不同；ensure_non_empty=False 避免空输出时无限递归
+            chunk_infer = replace(params_infer_code, manual_seed=42 + idx)
+            wavs = None
+            last_exc: Optional[BaseException] = None
+            for attempt in range(3):
+                try:
+                    wavs = chat.infer(
+                        chunk,
+                        skip_refine_text=False,
+                        params_refine_text=params_refine_text,
+                        params_infer_code=chunk_infer,
+                    )
+                    break
+                except RecursionError as exc:
+                    last_exc = exc
+                    chunk_infer.manual_seed = 1000 + idx * 10 + attempt
+                    release_cuda_cache()
+            if wavs is None:
+                return (
+                    False,
+                    f"ChatTTS 第 {idx}/{len(chunks)} 段合成失败（递归重试耗尽）。"
+                    f"请检查音色转写是否填写，或缩短该段文本。"
+                    f" 详情: {last_exc}",
+                    None,
+                )
             if not wavs or len(wavs) == 0:
                 return (
                     False,
@@ -670,7 +695,15 @@ def generate_voice(refined_text: str) -> Tuple[bool, str, Optional[str]]:
                     f"（段内容前 40 字: {chunk[:40]}…）",
                     None,
                 )
-            segment_wavs.append(np.asarray(wavs[0], dtype=np.float32))
+            wav_arr = np.asarray(wavs[0], dtype=np.float32)
+            if wav_arr.size == 0 or np.max(np.abs(wav_arr)) < 1e-6:
+                return (
+                    False,
+                    f"ChatTTS 第 {idx}/{len(chunks)} 段生成了空音频。"
+                    "请重新锁定音色并填写参考转写，或缩短润色稿后重试。",
+                    None,
+                )
+            segment_wavs.append(wav_arr)
             release_cuda_cache()
 
         wav_array = (
@@ -712,6 +745,13 @@ def generate_voice(refined_text: str) -> Tuple[bool, str, Optional[str]]:
                 "4. 确认无其他程序占用 GPU: nvidia-smi\n"
                 f"技术详情: {exc_msg[:400]}"
             )
+        elif "recursion depth" in exc_msg.lower() or isinstance(exc, RecursionError):
+            err = (
+                "语音合成失败: ChatTTS 反复生成空结果导致递归超限。\n"
+                "常见原因：未填写参考音频转写、润色稿含特殊符号、或音色文件异常。\n"
+                "处理：重新锁定音色并填写转写 → 用较短纯文本试合成。\n"
+                f"技术详情: {exc_msg[:400]}"
+            )
         elif "Corrupt input data" in exc_msg:
             err = (
                 "语音合成失败: 音色数据损坏或格式不兼容（Corrupt input data）。\n"