update

.gitignore

@@ -61,4 +61,5 @@ target/
 videos/
 logs/
 .idea/
+.DS_Store
 

config.py

@@ -9,8 +9,8 @@ class BaseConfig(object):
         'get_result': '/getResult'
     }
     XFASR_CONFIG = {
-        'appid': 'ac4ec700',
-        'secret_key': 'f822c63011275bcd26fa286fbb01768a'
+        'appid': 'b11df2ff',
+        'secret_key': '598f38c17bd361f1f473dcfc840d330a'
     }
 
     # gpt配置
@@ -39,6 +39,7 @@ class BaseConfig(object):
     TAGS_NEW = ['资讯', '国际', '政治', '科技', '军事', '健康', '社会', '民生政策', '人文', '人物', '旅行', '生活',
                 '美食', '综艺', '搞笑', '民俗文化', '安全', '休闲娱乐', '艺术', '情感']
     # GPT prompt
+    format_json_array = '[{},{},{}]'
     GPT_PROMPT = {
         'tags': {
             'prompt1': f"""
@@ -60,11 +61,14 @@ class BaseConfig(object):
 以json格式返回，key为categor与confidence, 分别代表类别与分类置信度。给出top 3的分类结果。
 -----------------------------
 """,
-            'prompt4': f"""
-请对如下文本进行分类。类别为其中的一个：【{' '.join(TAGS_NEW)}】。
-以json格式返回，key为category与confidence，分别代表类别与分类置信度。给出top 3的分类结果。
+            'prompt4': f"""请对如下文本进行分类。类别为其中的一个：【{' '.join(TAGS_NEW)}】。
+以json array格式返回，key为category与confidence，分别代表类别与分类置信度。给出top 3的分类结果。
 -----------------------------
 """,
+            'prompt5': f"""请对如下文本进行分类。类别为其中的一个：【{' '.join(TAGS_NEW)}】。
+以json array格式返回,{format_json_array},key为category与confidence，分别代表类别与分类置信度。给出top 3的分类结果。
+-----------------------------
+    """,
         },
         'title': {
             'prompt1': f"""

gpt_tag.py

@@ -10,6 +10,7 @@ log_ = Log()
 
 def get_tag(prompt):
     retry = 1
+    result_content = None
     while retry < 3:
         try:
             log_.info(f"retry = {retry}")
@@ -40,3 +41,4 @@ def get_tag(prompt):
             retry += 1
             time.sleep(30)
             continue
+    return result_content

main_process.py

@@ -1,3 +1,4 @@
+import json
 import traceback
 
 from feishu import FeiShuHelper
@@ -46,79 +47,114 @@ def main(sheet_info_config):
             # 1. 下载视频
             video_id = video['videoId']
             video_path = video['videoPath']
-            video_file = download_video(video_path=video_path, video_id=video_id, download_folder='videos')
-            # print(video_file)
-            log_.info(f"video_path = {video_file}")
-
-            # 2. 获取视频中的音频
-            audio_path = get_wav(video_path=video_file)
-            # print(audio_path)
-            log_.info(f"audio_path = {audio_path}")
-
-            # 3. asr
-            dialogue_path, asr_res_initial = call_asr(audio_path=audio_path)
-            # print(asr_res)
-            log_.info(f"asr_res_initial = {asr_res_initial}")
-
-            # 4. 判断asr识别的文本是否有效
-            validity = asr_validity_discrimination(text=asr_res_initial)
-            log_.info(f"validity = {validity}")
-
-            # 5. 对asr结果进行清洗
-            asr_res = asr_res_initial.strip().replace('\n', '')
-            for stop_word in config_.STOP_WORDS:
-                asr_res = asr_res.replace(stop_word, '')
-            # token限制: 字数 <= 2500
-            asr_res = asr_res[-2500:]
-
-            # 6. gpt产出结果
-            prompt = f"{config_.GPT_PROMPT['tags']['prompt4']}{asr_res.strip()}"
-            gpt_res = get_tag(prompt=prompt)
-            # print(gpt_res)
-            log_.info(f"gpt_res = {gpt_res}")
-
-            # 7. 结果写入飞书表格
-            result = [[video_id, video_path, video['title'], str(validity), asr_res_initial, gpt_res, prompt]]
-            log_.info(f"result = {result}")
-            if len(result) > 0:
-                feishu_helper.data_to_feishu_sheet(
-                    sheet_token=res_spreadsheet_token,
-                    sheet_id=res_sheet_id,
-                    data=result,
-                    start_row=write_start_row,
-                    start_column=write_start_col,
-                    end_column=write_end_col
-                )
-                log_.info(f"write to feishu success!")
-                write_start_row += 1
+            if video_path[-4:] != '.mp4':
+                result = [[video_id, video_path, video['title'], '', '', '', '', '']]
+                log_.info(f"result = {result}")
+                if len(result) > 0:
+                    feishu_helper.data_to_feishu_sheet(
+                        sheet_token=res_spreadsheet_token,
+                        sheet_id=res_sheet_id,
+                        data=result,
+                        start_row=write_start_row,
+                        start_column=write_start_col,
+                        end_column=write_end_col
+                    )
+                    log_.info(f"write to feishu success!")
+                    write_start_row += 1
+            else:
+                try:
+                    video_file = download_video(video_path=video_path, video_id=video_id, download_folder='videos')
+                    # print(video_file)
+                    log_.info(f"video_path = {video_file}")
+
+                    # 2. 获取视频中的音频
+                    audio_path = get_wav(video_path=video_file)
+                    # print(audio_path)
+                    log_.info(f"audio_path = {audio_path}")
+
+                    # 3. asr
+                    dialogue_path, asr_res_initial = call_asr(audio_path=audio_path)
+                    # print(asr_res)
+                    log_.info(f"asr_res_initial = {asr_res_initial}")
+                except:
+                    log_.error(traceback.format_exc())
+                    result = [[video_id, video_path, video['title'], '', '', '', '', '']]
+                    log_.info(f"result = {result}")
+                    if len(result) > 0:
+                        feishu_helper.data_to_feishu_sheet(
+                            sheet_token=res_spreadsheet_token,
+                            sheet_id=res_sheet_id,
+                            data=result,
+                            start_row=write_start_row,
+                            start_column=write_start_col,
+                            end_column=write_end_col
+                        )
+                        log_.info(f"write to feishu success!")
+                        write_start_row += 1
+                    continue
+
+                # 4. 判断asr识别的文本是否有效
+                validity = asr_validity_discrimination(text=asr_res_initial)
+                log_.info(f"validity = {validity}")
+                if validity is True:
+                    # 5. 对asr结果进行清洗
+                    asr_res = asr_res_initial.strip().replace('\n', '')
+                    for stop_word in config_.STOP_WORDS:
+                        asr_res = asr_res.replace(stop_word, '')
+                    # token限制: 字数 <= 2500
+                    asr_res = asr_res[-2500:]
+
+                    # 6. gpt产出结果
+                    prompt = f"{config_.GPT_PROMPT['tags']['prompt5']}{asr_res.strip()}"
+                    gpt_res = get_tag(prompt=prompt)
+                    # print(gpt_res)
+                    log_.info(f"gpt_res = {gpt_res}, type = {type(gpt_res)}")
+
+                    # 7. 结果写入飞书表格
+                    if gpt_res is None:
+                        result = [[video_id, video_path, video['title'], str(validity), asr_res_initial, prompt, '', '']]
+                    else:
+                        confidence_up_list = []
+                        try:
+                            for item in json.loads(gpt_res):
+                                if item['confidence'] > 0.5:
+                                    confidence_up_list.append(item['category'])
+                        except:
+                            pass
+                        confidence_up = ', '.join(confidence_up_list)
+                        result = [[video_id, video_path, video['title'], str(validity), asr_res_initial,
+                                   prompt, gpt_res, confidence_up]]
+                else:
+                    result = [[video_id, video_path, video['title'], str(validity), asr_res_initial, '', '', '']]
+                log_.info(f"result = {result}")
+                if len(result) > 0:
+                    feishu_helper.data_to_feishu_sheet(
+                        sheet_token=res_spreadsheet_token,
+                        sheet_id=res_sheet_id,
+                        data=result,
+                        start_row=write_start_row,
+                        start_column=write_start_col,
+                        end_column=write_end_col
+                    )
+                    log_.info(f"write to feishu success!")
+                    write_start_row += 1
         except Exception as e:
             log_.error(e)
             log_.error(traceback.format_exc())
             continue
 
-    # 6. 结果写入飞书表格
-    # if len(result) > 0:
-    #     feishu_helper.data_to_feishu_sheet(
-    #         sheet_token=res_spreadsheet_token,
-    #         sheet_id=res_sheet_id,
-    #         data=result,
-    #         start_row=write_start_row,
-    #         start_column=write_start_col,
-    #         end_column=write_end_col
-    #     )
-
 
 if __name__ == '__main__':
     sheet_info = {
         '历史视频top5000回流倒叙排列': {
             'video_spreadsheet_token': 'L4ywsRaV2hFLv1t4Athcdw71nde',
             'video_sheet_id': 'hRjMrL',
-            'read_start_row': 2,
+            'read_start_row': 44,
             'res_spreadsheet_token': 'DkiUsqwJ6hmBxstBYyEcNE4ante',
             'res_sheet_id': '7Fua00',
-            'write_start_row': 3,
+            'write_start_row': 44,
             'write_start_col': 'A',
-            'write_end_col': 'I'
+            'write_end_col': 'H'
         }
     }
 

xunfei_asr.py

@@ -79,6 +79,7 @@ class RequestApi(object):
             headers={"Content-type": "application/json"},
             data=data
         )
+        # print(response.text)
         # print("upload_url:", response.request.url)
         result = json.loads(response.text)
         # print("upload resp:", result)
@@ -150,5 +151,5 @@ def call_asr(audio_path):
 
 
 if __name__ == '__main__':
-    audio_path = 'videos/001.wav'
+    audio_path = 'videos/1275943.wav'
     call_asr(audio_path=audio_path)