import json import logging import time import uuid import oss2 from aliyunsdkcore.client import AcsClient from aliyunsdkcore.acs_exception.exceptions import ClientException, ServerException # 尝试导入最新的 API 版本，如果有问题可能需要调整 try: from aliyunsdktingwu.request.v20230930 import CreateTaskRequest, GetTaskInfoRequest except ImportError: # Fallback or error handling if version differs pass from django.conf import settings logger = logging.getLogger(__name__) from .models import TranscriptionTask, AIEvaluation, AIEvaluationTemplate class AliyunTingwuService: def __init__(self): self.access_key_id = settings.ALIYUN_ACCESS_KEY_ID self.access_key_secret = settings.ALIYUN_ACCESS_KEY_SECRET self.oss_bucket_name = settings.ALIYUN_OSS_BUCKET_NAME self.oss_endpoint = settings.ALIYUN_OSS_ENDPOINT self.tingwu_app_key = settings.ALIYUN_TINGWU_APP_KEY self.region_id = "cn-shanghai" # 听悟服务区域，根据文档应与OSS区域一致，或者使用 'cn-beijing' # 初始化 OSS Bucket if self.access_key_id and self.access_key_secret and self.oss_endpoint: auth = oss2.Auth(self.access_key_id, self.access_key_secret) self.bucket = oss2.Bucket(auth, self.oss_endpoint, self.oss_bucket_name) else: self.bucket = None logger.warning("Aliyun OSS configuration missing.") # 初始化听悟 Client if self.access_key_id and self.access_key_secret: self.client = AcsClient( self.access_key_id, self.access_key_secret, self.region_id ) # 显式添加听悟服务的 Endpoint 映射，解决 EndpointResolvingError # 听悟 API 的服务接入点通常是 tingwu.cn-beijing.aliyuncs.com # 但新版听悟 API (tingwu.aliyuncs.com) 可能不同，需根据实际情况添加 # 这里添加一个通用的 Endpoint 映射 try: # 尝试为 tingwu 产品设置 Endpoint # 注意：听悟服务主要部署在北京，Endpoint 通常为 tingwu.cn-beijing.aliyuncs.com # 如果您的服务在上海，也可能需要连接到北京的接入点 self.client.add_endpoint(self.region_id, "tingwu", "tingwu.cn-beijing.aliyuncs.com") except Exception as e: logger.warning(f"Failed to add endpoint: {e}") else: self.client = None logger.warning("Aliyun AccessKey configuration missing.") def upload_to_oss(self, file_obj, file_name, day=7): """ 上传文件到 OSS 并返回带签名的 URL 默认生成有效期为 7 天 (3600 * 24 * day) 的签名URL，方便评委在一段时间内都能播放。 """ if not self.bucket: raise Exception("OSS Client not initialized") try: # 上传文件 # file_obj 应该是打开的文件对象或字节流 self.bucket.put_object(file_name, file_obj) # 生成签名 URL，有效期 7 天 (3600 * 24 * 7 = 604800 秒) url = self.bucket.sign_url('GET', file_name, 3600 * 24 * day) return url except Exception as e: logger.error(f"OSS Upload failed: {e}") raise e def create_transcription_task(self, file_url, language="cn"): """ 创建听悟转写任务 """ if not self.client: raise Exception("Tingwu Client not initialized") request = CreateTaskRequest.CreateTaskRequest() # 针对阿里云 SDK 不同版本的兼容性处理 # "type" 参数是听悟 API (ROA 风格) 的必填项，用于指定任务类型 # 根据官方文档，离线任务的 type 通常就是 'offline' request.add_query_param('type', 'offline') # 构造请求体 (Body) # 根据听悟 API 文档，AppKey, Input, Parameters 应位于 JSON Body 中 # 而不是 Query Parameter body = { "AppKey": self.tingwu_app_key, "Input": { "FileUrl": file_url, "SourceLanguage": language, "TaskKey": str(uuid.uuid4()) }, "Parameters": { "Transcoding": { "TargetAudioFormat": "mp3" }, "Transcription": { "DiarizationEnabled": True, "ChannelId": 0 }, "TranscriptionEnabled": True, "AutoChaptersEnabled": True, "SummarizationEnabled": True, "Summarization": { "Types": ["Paragraph", "Conversational", "QuestionsAnswering", "MindMap"] } } } # 设置 Body 内容 request.set_content(json.dumps(body)) request.add_header('Content-Type', 'application/json') # 强制设置 Endpoint，避免 SDK.EndpointResolvingError # 听悟目前主要服务点在北京 request.set_endpoint("tingwu.cn-beijing.aliyuncs.com") # 显式设置 Method 为 PUT request.set_method('PUT') try: response = self.client.do_action_with_exception(request) return json.loads(response) except (ClientException, ServerException) as e: logger.error(f"Tingwu CreateTask failed: {e}") raise e def get_task_info(self, task_id): """ 查询任务状态和结果 """ if not self.client: raise Exception("Tingwu Client not initialized") request = GetTaskInfoRequest.GetTaskInfoRequest() request.set_TaskId(task_id) try: response = self.client.do_action_with_exception(request) return json.loads(response) except (ClientException, ServerException) as e: logger.error(f"Tingwu GetTaskInfo failed: {e}") raise e def parse_and_update_task(self, task, result): """ 解析听悟结果并更新任务 :param task: TranscriptionTask 实例 :param result: get_task_info 返回的完整 JSON (或 Data 部分) """ # 记录之前的状态，用于判断是否是首次完成 previous_status = task.status # 1. 提取 Data 对象 if isinstance(result, dict): data_obj = result.get('Data', result) else: data_obj = result if not isinstance(data_obj, dict): logger.error(f"Unexpected data format: {type(data_obj)}") return # 2. 更新状态 task_status = data_obj.get('TaskStatus') or data_obj.get('Status') if task_status in ['COMPLETE', 'COMPLETED', 'SUCCEEDED']: task.status = 'SUCCEEDED' # 使用字符串引用，避免导入模型循环引用 elif task_status == 'FAILED': task.status = 'FAILED' task.error_message = data_obj.get('TaskStatusText', data_obj.get('Message', 'Unknown error')) task.save() return else: # 仍在处理中，不更新内容 return # 3. 解析结果 task_result = data_obj.get('Result', {}) # 兼容处理：如果 Result 为空，或者不存在，尝试直接使用 data_obj 作为结果源 # 某些情况下，Summarization/AutoChapters 可能直接位于 Data 层级 if not task_result: task_result = data_obj # 辅助函数：从源字典或其 Result 子字典中获取字段 def get_data_field(source, key): # 1. 尝试直接从 task_result 获取 (如果 task_result 就是 Data 本身，这里也会生效) if isinstance(source, dict) and key in source: return source[key] # 2. 如果 source 是 Data，尝试从 source['Result'] 获取 if isinstance(source, dict) and 'Result' in source and isinstance(source['Result'], dict): if key in source['Result']: return source['Result'][key] return None # --- A. 处理逐字稿 (Transcription) --- transcription_data = get_data_field(task_result, 'Transcription') or get_data_field(data_obj, 'Transcription') or {} # 处理 URL 下载 if isinstance(transcription_data, str) and transcription_data.startswith('http'): try: import requests t_resp = requests.get(transcription_data) if t_resp.status_code == 200: transcription_data = t_resp.json() except Exception as e: logger.error(f"Download transcription failed: {e}") transcription_data = {} elif isinstance(transcription_data, dict) and 'TranscriptionUrl' in transcription_data: try: import requests t_resp = requests.get(transcription_data['TranscriptionUrl']) if t_resp.status_code == 200: transcription_data = t_resp.json() except Exception as e: logger.error(f"Download transcription url failed: {e}") # 保存原始数据 task.transcription_data = transcription_data # 提取文本 # 结构: {"Transcription": {"Paragraphs": [{"Words": [{"Text": "..."}]}]}} # 或直接 {"Paragraphs": ...} content_source = transcription_data if 'Transcription' in content_source and isinstance(content_source['Transcription'], dict): content_source = content_source['Transcription'] paragraphs = content_source.get('Paragraphs', []) full_text_lines = [] if paragraphs and isinstance(paragraphs, list): for p in paragraphs: # 尝试从 Words 中提取 words = p.get('Words', []) if words: line_text = "".join([str(w.get('Text', '')) for w in words]) full_text_lines.append(line_text) # 兼容旧结构或直接 Text elif 'Text' in p: full_text_lines.append(p['Text']) if full_text_lines: task.transcription = "\n".join(full_text_lines) # --- B. 处理 AI 总结 (Summarization) --- summarization = get_data_field(task_result, 'Summarization') or get_data_field(data_obj, 'Summarization') or {} # 处理 URL 下载 if isinstance(summarization, str) and summarization.startswith('http'): try: import requests s_resp = requests.get(summarization) if s_resp.status_code == 200: summarization = s_resp.json() except Exception as e: logger.error(f"Download summarization failed: {e}") summarization = {} # 保存原始数据 task.summary_data = summarization # 提取文本 (MindMapSummary) # 结构: {"MindMapSummary": [{"Title": "...", "Topic": [...]}]} # 移除了原先的 summary_text 拼接逻辑 # --- C. 处理章节 (AutoChapters) --- auto_chapters = get_data_field(task_result, 'AutoChapters') or get_data_field(data_obj, 'AutoChapters') or [] # 处理 URL 下载 if isinstance(auto_chapters, str) and auto_chapters.startswith('http'): try: import requests ac_resp = requests.get(auto_chapters) if ac_resp.status_code == 200: auto_chapters = ac_resp.json() except Exception as e: logger.error(f"Download auto chapters failed: {e}") auto_chapters = [] # 保存原始数据 task.auto_chapters_data = auto_chapters # 保存任务，确保原始数据已写入数据库 task.save() # 调用大模型生成总结 (如果 summary_data 或 auto_chapters_data 存在) if task.summary_data or task.auto_chapters_data: try: # 设置占位状态 task.summary = "AI总结生成当中..." task.save(update_fields=['summary']) # 异步执行总结 import threading from .bailian_service import BailianService def async_summarize_in_service(task_id): try: # 重新获取 task 以避免线程安全问题 from .models import TranscriptionTask t = TranscriptionTask.objects.get(id=task_id) bailian_service = BailianService() bailian_service.summarize_task(t) except Exception as e: logger.error(f"Async summary generation failed in service: {e}") threading.Thread(target=async_summarize_in_service, args=(task.id,)).start() logger.info(f"Triggered async summary generation for task {task.id}") except Exception as e: logger.error(f"Failed to trigger AI summarization: {e}") # 4. 自动触发 AI 评估 (如果任务首次成功且有启用的模板) if previous_status != 'SUCCEEDED' and task.status == 'SUCCEEDED' and task.transcription: # 同样改为异步触发，传递 task.id 以避免线程中的对象状态问题 import threading threading.Thread(target=self.trigger_ai_evaluations, args=(task.id,)).start() def trigger_ai_evaluations(self, task_id): """ 根据启用的模板自动触发 AI 评估 逻辑： 1. 如果模板关联了评分维度(s score_dimension)，只对关联了相同维度的比赛进行评估 2. 如果模板未关联评分维度： - 如果是默认模板(is_default=True)，评价所有比赛 - 否则不进行自动评价 """ try: # 在线程中重新获取 task 对象，并预加载 project，避免懒加载导致的线程数据库连接问题 from .models import TranscriptionTask task = TranscriptionTask.objects.select_related('project', 'project__competition').get(id=task_id) except Exception as e: # 兼容处理：如果 task_id 其实是 task 对象（虽然我们上面改了，但防止其他地方调用传错） if hasattr(task_id, 'id'): try: from .models import TranscriptionTask task = TranscriptionTask.objects.select_related('project', 'project__competition').get(id=task_id.id) except: task = task_id else: logger.error(f"Failed to retrieve task {task_id}: {e}") return active_templates = AIEvaluationTemplate.objects.filter(is_active=True) if not active_templates.exists(): logger.info("No active AI evaluation templates found.") return from .bailian_service import BailianService service = BailianService() for template in active_templates: # 检查是否已经存在相同的评估，避免重复创建 if AIEvaluation.objects.filter(task=task, template=template).exists(): logger.info(f"Evaluation for task {task.id} and template {template.name} already exists.") continue # 获取任务关联的比赛 task_competition = None if task.project and task.project.competition: task_competition = task.project.competition # 判断是否应该对此任务进行评估 should_evaluate = False if template.score_dimension: # 模板关联了评分维度，只对关联了相同维度的比赛进行评估 if task_competition: # 获取该比赛下所有关联了相同评分维度的比赛ID列表 from competition.models import ScoreDimension related_competition_ids = ScoreDimension.objects.filter( id=template.score_dimension.id ).values_list('competition_id', flat=True) if task_competition.id in related_competition_ids: should_evaluate = True logger.info(f"Template '{template.name}' is linked to score_dimension, task's competition matches.") else: logger.info(f"Template '{template.name}' is linked to score_dimension, but task's competition does not match. Skipping.") else: logger.info(f"Task {task.id} has no associated competition. Skipping template '{template.name}'.") else: # 模板未关联评分维度，只有默认模板才评价所有比赛 if template.is_default: should_evaluate = True logger.info(f"Template '{template.name}' is default template, evaluating all competitions.") else: logger.info(f"Template '{template.name}' is not linked to score_dimension and is not default. Skipping.") if not should_evaluate: continue # 创建评估记录 evaluation = AIEvaluation.objects.create( task=task, template=template, model_selection=template.model_selection, prompt=template.prompt, status=AIEvaluation.Status.PENDING ) # 触发评估 try: service.evaluate_task(evaluation) logger.info(f"Triggered evaluation {evaluation.id} for template {template.name}") except Exception as e: logger.error(f"Failed to trigger evaluation {evaluation.id}: {e}")