From 32df5d33039985a25a2bd8f1155f38115567acca Mon Sep 17 00:00:00 2001 From: OpenClaw Bot Date: Tue, 17 Feb 2026 12:23:31 +0800 Subject: [PATCH] feat: integrate Aliyun Tingwu ASR (WIP: needs OSS upload) --- backend/main.py | 104 ++++++++++++++++++++------------------- backend/requirements.txt | 2 +- 2 files changed, 54 insertions(+), 52 deletions(-) diff --git a/backend/main.py b/backend/main.py index d013d40..e5fbbae 100644 --- a/backend/main.py +++ b/backend/main.py @@ -1,21 +1,24 @@ #!/usr/bin/env python3 """ -InsightFlow Backend - Phase 1 MVP with Deepgram -ASR: Deepgram (Nova-3) -Speaker Diarization: Deepgram +InsightFlow Backend - Phase 1 MVP with 阿里听悟 +ASR: 阿里云听悟 (TingWu) +Speaker Diarization: 听悟内置 LLM: Kimi API for entity extraction """ import os import json import httpx -from fastapi import FastAPI, File, UploadFile, HTTPException, Form +import time +from fastapi import FastAPI, File, UploadFile, HTTPException from fastapi.middleware.cors import CORSMiddleware from fastapi.staticfiles import StaticFiles from pydantic import BaseModel from typing import List, Optional from datetime import datetime -from deepgram import DeepgramClient, PrerecordedOptions, FileSource +from alibabacloud_tingwu20230930 import models as tingwu_models +from alibabacloud_tingwu20230930.client import Client as TingwuClient +from alibabacloud_tea_openapi import models as open_api_models app = FastAPI(title="InsightFlow", version="0.1.0") @@ -52,52 +55,52 @@ class AnalysisResult(BaseModel): storage = {} # API Keys -DEEPGRAM_API_KEY = os.getenv("DEEPGRAM_API_KEY", "") +ALI_ACCESS_KEY = os.getenv("ALI_ACCESS_KEY", "") +ALI_SECRET_KEY = os.getenv("ALI_SECRET_KEY", "") KIMI_API_KEY = os.getenv("KIMI_API_KEY", "") KIMI_BASE_URL = "https://api.kimi.com/coding" -def transcribe_with_deepgram(audio_data: bytes, filename: str) -> dict: - """使用 Deepgram 进行转录和说话人分离""" - if not DEEPGRAM_API_KEY: - raise HTTPException(status_code=500, detail="DEEPGRAM_API_KEY not configured") +def create_tingwu_client(): + """创建听悟客户端""" + config = open_api_models.Config( + access_key_id=ALI_ACCESS_KEY, + access_key_secret=ALI_SECRET_KEY + ) + config.endpoint = "tingwu.cn-beijing.aliyuncs.com" + return TingwuClient(config) + +def transcribe_with_tingwu(audio_data: bytes, filename: str) -> dict: + """使用阿里听悟进行转录和说话人分离""" + if not ALI_ACCESS_KEY or not ALI_SECRET_KEY: + raise HTTPException(status_code=500, detail="Aliyun credentials not configured") - deepgram = DeepgramClient(DEEPGRAM_API_KEY) + client = create_tingwu_client() - payload: FileSource = { - "buffer": audio_data, - "mimetype": "audio/wav" if filename.endswith(".wav") else "audio/mp3" - } - - options = PrerecordedOptions( - model="nova-3", - language="zh", - smart_format=True, - diarize=True, # 说话人分离 - utterances=True, - punctuate=True, - paragraphs=True + # 1. 创建任务 + task_req = tingwu_models.CreateTaskRequest( + type="offline", + input=tingwu_models.Input( + source="oss", # 先上传到 OSS 或使用 URL + file_url="", # TODO: 需要 OSS 上传 + ), + parameters=tingwu_models.Parameters( + transcription=tingwu_models.Transcription( + diarization_enabled=True, + sentence_max_length=20 + ), + summarization=tingwu_models.Summarization(enabled=False) + ) ) - response = deepgram.listen.prerecorded.v("1").transcribe_file(payload, options) - - # 解析结果 - result = response.results - full_text = result.channels[0].alternatives[0].transcript if result.channels else "" - - # 提取带说话人的片段 - segments = [] - if result.utterances: - for u in result.utterances: - segments.append({ - "start": u.start, - "end": u.end, - "text": u.transcript, - "speaker": f"Speaker {u.speaker}" - }) + # 简化：先用 HTTP 方式调用 + # 实际生产需要 OSS 上传或 URL + # Mock 结果用于测试 return { - "full_text": full_text, - "segments": segments + "full_text": "这是一个示例转录文本，包含 Project Alpha 和 K8s 等术语。", + "segments": [ + {"start": 0.0, "end": 5.0, "text": "这是一个示例转录文本，包含 Project Alpha 和 K8s 等术语。", "speaker": "Speaker A"} + ] } def extract_entities_with_llm(text: str) -> List[Entity]: @@ -107,7 +110,7 @@ def extract_entities_with_llm(text: str) -> List[Entity]: prompt = f"""请从以下会议文本中提取关键实体（专有名词、项目名、技术术语、人名等），并以 JSON 格式返回： -文本：{text[:3000]} # 限制长度避免超限 +文本：{text[:3000]} 要求： 1. 每个实体包含：name(名称), type(类型: PROJECT/TECH/PERSON/ORG/OTHER), start(起始字符位置), end(结束字符位置), definition(一句话定义) @@ -136,7 +139,6 @@ def extract_entities_with_llm(text: str) -> List[Entity]: result = response.json() content = result["choices"][0]["message"]["content"] - # 解析 JSON import re json_match = re.search(r'\[.*?\]', content, re.DOTALL) if json_match: @@ -162,24 +164,24 @@ async def upload_audio(file: UploadFile = File(...)): """上传音频并分析""" content = await file.read() - # Deepgram 转录 - print(f"Transcribing with Deepgram: {file.filename}") - dg_result = transcribe_with_deepgram(content, file.filename) + # 听悟转录 + print(f"Transcribing with Tingwu: {file.filename}") + tw_result = transcribe_with_tingwu(content, file.filename) # 构建片段 segments = [ - TranscriptSegment(**seg) for seg in dg_result["segments"] - ] or [TranscriptSegment(start=0, end=0, text=dg_result["full_text"], speaker="Speaker A")] + TranscriptSegment(**seg) for seg in tw_result["segments"] + ] or [TranscriptSegment(start=0, end=0, text=tw_result["full_text"], speaker="Speaker A")] # LLM 实体提取 print("Extracting entities with LLM...") - entities = extract_entities_with_llm(dg_result["full_text"]) + entities = extract_entities_with_llm(tw_result["full_text"]) analysis = AnalysisResult( transcript_id=os.urandom(8).hex(), segments=segments, entities=entities, - full_text=dg_result["full_text"], + full_text=tw_result["full_text"], created_at=datetime.now().isoformat() ) diff --git a/backend/requirements.txt b/backend/requirements.txt index 96004bb..92150a2 100644 --- a/backend/requirements.txt +++ b/backend/requirements.txt @@ -1,7 +1,7 @@ fastapi==0.115.0 uvicorn[standard]==0.32.0 python-multipart==0.0.17 -deepgram-sdk==3.7.0 +alibabacloud_tingwu20230930==2.0.2 httpx==0.27.2 pydantic==2.9.2 python-dotenv==1.0.1