最新哦，分词策略放到cli了

src/agent/llm.py

@@ -10,6 +10,7 @@ from pydantic import Field
 import re, json
 from datetime import datetime as dt
 import requests
+from concurrent.futures import ThreadPoolExecutor, as_completed
 
 from langchain_core.messages import HumanMessage, AIMessage, BaseMessage, SystemMessage  
 from langchain_core.outputs import ChatResult, ChatGeneration
@@ -17,7 +18,7 @@ from langchain_openai import ChatOpenAI
 
 from agent.utils.image_utils import image_bytes_to_openai_image_url_part
 from agent.utils.env_utils import LLM_API_KEY, LLM_BASE_URL, LLM_MODEL, LLM_TIMEOUT, TTS_BASE_URL, ASR_BASE_URL
-from agent.utils.agent_utils import assistant_text, Documents_process
+from agent.utils.agent_utils import assistant_text, Documents_process, split_text, merge_wavs
 from agent.utils.image_utils import image_path_to_openai_image_url_part, image_bytes_to_openai_image_url_part, IMAGE_FILE_EXTENSIONS
 from agent.memory.system_prompt import VISION_SYSTEM_PROMPT
 
@@ -218,9 +219,30 @@ class ChatQwen(ChatOpenAI):
         return new_messages
 
     def _text_to_audio(self, text: str) -> bytes:
-        resp = requests.get(f"{TTS_BASE_URL}/tts", params={"text": text})
-        resp.raise_for_status()
-        return resp.content
+        chunks = split_text(text)
+        if len(chunks) == 1:
+            resp = requests.get(f"{TTS_BASE_URL}/tts", params={"text": chunks[0]})
+            resp.raise_for_status()
+            return resp.content
+
+        # 多块并行请求
+        audio_parts = [None] * len(chunks)
+        with ThreadPoolExecutor(max_workers=min(len(chunks), 4)) as pool:
+            future_to_idx = {
+                pool.submit(
+                    lambda t: requests.get(
+                        f"{TTS_BASE_URL}/tts", params={"text": t}
+                    ).content,
+                    chunk
+                ): idx for idx, chunk in enumerate(chunks)
+            }
+            for future in as_completed(future_to_idx):
+                idx = future_to_idx[future]
+                try:
+                    audio_parts[idx] = future.result()
+                except Exception as e:
+                    raise RuntimeError(f"TTS request failed for chunk {idx}: {e}")
+        return merge_wavs(audio_parts)
 
     @staticmethod
     def _get_wav_duration_seconds(audio_bytes: bytes) -> float | None:

src/agent/utils/agent_utils.py

@@ -5,7 +5,7 @@ import shutil
 import sqlite3
 from typing import List, Tuple, Any
 import pathlib
-import mimetypes
+import mimetypes, re
 
 from langchain_community.chat_message_histories import SQLChatMessageHistory
 from langchain_core.messages import BaseMessage, SystemMessage, HumanMessage
@@ -527,32 +527,50 @@ class Documents_process:
             return Documents_process.process_tabular(file_path)
         else:
             raise ValueError(f"未处理的文件类型: {file_type}")
+# 文本分块工具
+SENT_SPLIT_PATTERN = re.compile(r'(?<=[。！？；，、])')
+REMOVE_PUNCT_PATTERN = re.compile(r'[。！？；，、：；“”（）《》【】\-\_\—\~\.\,\!\?\s·]')
+def effective_length(text: str) -> int:
+    return len(REMOVE_PUNCT_PATTERN.sub('', text))
 
-if __name__ == "__main__":
-    docx_path = r"C:\Users\86166\Desktop\Agent_Learning_minor\信息科学与工程学院-赵永顺-《兼爱》——墨翟一生的“白月光”.docx"
-    pdf_path = r"C:\Users\86166\Desktop\Agent_Learning_minor\信息科学与工程学院-赵永顺-《兼爱》——墨翟一生的“白月光”.pdf"
-    xlsx_path = r"C:\Users\86166\Desktop\Agent_Learning_minor\电科二班2025德育评价.xlsx"
-    py_path = r"C:\Users\86166\Desktop\Agent_Learning_minor\speech_server.py"
+def split_text(text: str, min_effective_chars: int = 20) -> list[str]:
+    # 先按标点粗分
+    raw_chunks = [c.strip() for c in SENT_SPLIT_PATTERN.split(text) if c.strip()]
+    if not raw_chunks:
+        return [text]
+
+    merged = []
+    buffer = ""
+    for chunk in raw_chunks:
+        buffer += chunk
+        # 当累积的有效字符数达到阈值，或已是最后一个块时，切出一块
+        if effective_length(buffer) >= min_effective_chars:
+            merged.append(buffer)
+            buffer = ""
+    if buffer:  # 剩余部分（必然小于阈值）
+        if merged:
+            merged[-1] += buffer   # 合并到最后一块
+        else:
+            merged.append(buffer)  # 整段文本都太短，保留原样
+    return merged
     
-    output_path = "test_results.txt"
-    with open(output_path, "w", encoding="utf-8") as f:
-        test_files = [
-            ("py", py_path),
-            ("docx", docx_path),
-            ("xlsx", xlsx_path),
-            ("pdf", pdf_path),
-        ]
-        for label, path in test_files:
-            f.write(f"\n===== 测试 {label} 文件：{path} =====\n")
-            try:
-                text, images = Documents_process.process(path)
-                f.write(f"提取文本长度：{len(text)}\n")
-                f.write(f"提取图片数量：{len(images)}\n")
-                f.write("文本预览：\n")
-                f.write(text)
-            except Exception as e:
-                f.write(f"❌ 处理失败: {e}\n")
-    print(f"测试结果已写入 {output_path}")
-
+def merge_wavs(audio_parts: list[bytes]) -> bytes:
+    """将多个 WAV 字节串按顺序拼接成一个 WAV"""
+    if len(audio_parts) == 1:
+        return audio_parts[0]
+    import io, wave
+    data_frames = []
+    params = None
+    for part in audio_parts:
+        with wave.open(io.BytesIO(part), 'rb') as wf:
+            if params is None:
+                params = wf.getparams()
+            data_frames.append(wf.readframes(wf.getnframes()))
+    buf = io.BytesIO()
+    with wave.open(buf, 'wb') as wf_out:
+        wf_out.setparams(params)
+        for frames in data_frames:
+            wf_out.writeframes(frames)
+    return buf.getvalue()
 
     

src/web/server.py

@@ -27,21 +27,13 @@ if str(_SRC) not in sys.path:
 from agent import chatbot_agent as ca  # noqa: E402
 from web import ui_session as ui  # noqa: E402
 from agent.utils.image_utils import IMAGE_FILE_EXTENSIONS  # noqa: E402
+from agent.utils.agent_utils import DOC_EXT_MAP  # noqa: E402
 
 _WEB_DIR = Path(__file__).resolve().parent
 _IMAGE_DIR = _WEB_DIR / "image"
 
 _AUDIO_EXT = frozenset({".wav", ".mp3", ".m4a", ".flac", ".ogg", ".webm"})
-_IMAGE_EXT = frozenset(IMAGE_FILE_EXTENSIONS)
-
-
-def _infer_upload_kind(fp: str) -> str | None:
-    ext = os.path.splitext(fp)[1].lower()
-    if ext in _IMAGE_EXT:
-        return "image"
-    if ext in _AUDIO_EXT:
-        return "audio"
-    return "file"
+_FILE_EXT = frozenset(DOC_EXT_MAP.keys())
 
 
 def _data_url_for_file(fp: str) -> str | None:
@@ -92,7 +84,6 @@ def _serialize_content(content: Any, *, meta: dict[str, Any] | None = None, role
             parts.append({"type": "text", "text": piece.get("text") or ""})
             continue
         fp = piece.get("path") or (piece.get("file") or {}).get("path")
-        display_name = piece.get("name")
         if piece.get("type") == "image_url":
             fp = fp or (piece.get("image_url") or {}).get("url")
         if piece.get("type") == "audio":
@@ -104,19 +95,19 @@ def _serialize_content(content: Any, *, meta: dict[str, Any] | None = None, role
             continue
         ext = os.path.splitext(fp)[1].lower()
         if ext in IMAGE_FILE_EXTENSIONS:
-            parts.append({"type": "image", "url": data_url, "path": fp, "name": display_name or os.path.basename(fp)})
+            parts.append({"type": "image", "url": data_url, "path": fp, "name": os.path.basename(fp)})
         elif ext in _AUDIO_EXT:
             audio_meta = meta.get("audio") if isinstance(meta.get("audio"), dict) else {}
             parts.append({
                 "type": "audio",
                 "url": data_url,
                 "path": fp,
-                "name": display_name or os.path.basename(fp),
+                "name": os.path.basename(fp),
                 "autoplay": bool(meta.get("autoplay") or audio_meta.get("autoplay")) if role == "assistant" else False,
                 "duration_seconds": audio_meta.get("duration_seconds"),
             })
         else:
-            parts.append({"type": "file", "url": data_url, "path": fp, "name": display_name or os.path.basename(fp)})
+            parts.append({"type": "file", "url": data_url, "path": fp, "name": os.path.basename(fp)})
     return parts
 
 
@@ -253,7 +244,6 @@ def _materialize_uploads_to_temp(
     """将上传文件落盘到临时目录，返回 ``(query_dict, tmp_root)``；调用方须在 finally 中删除 ``tmp_root``。"""
     tmp_root = tempfile.mkdtemp(prefix="web_mm_")
     file_paths: list[str] = []
-    file_names: list[str] = []
     for uf in images or []:
         if uf is None or not uf.filename:
             continue
@@ -262,7 +252,6 @@ def _materialize_uploads_to_temp(
         with open(dest, "wb") as f:
             shutil.copyfileobj(uf.file, f)
         file_paths.append(dest)
-        file_names.append(uf.filename)
 
     for uf in files or []:
         if uf is None or not uf.filename:
@@ -272,7 +261,6 @@ def _materialize_uploads_to_temp(
         with open(dest, "wb") as f:
             shutil.copyfileobj(uf.file, f)
         file_paths.append(dest)
-        file_names.append(uf.filename)
 
     if audio is not None and audio.filename:
         ext = Path(audio.filename).suffix or ".bin"
@@ -280,9 +268,8 @@ def _materialize_uploads_to_temp(
         with open(dest, "wb") as f:
             shutil.copyfileobj(audio.file, f)
         file_paths.append(dest)
-        file_names.append(audio.filename)
 
-    query = {"text": (text or "").strip(), "files": file_paths, "file_names": file_names}
+    query = {"text": (text or "").strip(), "files": file_paths}
     return query, tmp_root