基本完成

.dockerignore

@@ -6,4 +6,3 @@ outputs
 tests
 
 **/checkpoints
-/app/ai_gen_image/images
\ No newline at end of file

Dockerfile

-FROM harbor.5jstore.com:8020/ai/wm_generate_ai:v0.2
+FROM harbor.5jstore.com:8020/ai/wm_video_cut:v0.2
 LABEL maintainer="zhouchengbo@wmdigit.com"
 
 ARG DEBIAN_FRONTEND=noninteractive

Dockerfile_v0.2

-FROM harbor.5jstore.com:8020/ai/wm_generate_ai:v0.1
+FROM harbor.5jstore.com:8020/ai/wm_video_cut:v0.1
 LABEL maintainer="zhouchengbo@wmdigit.com"
 
 ARG DEBIAN_FRONTEND=noninteractive

README.md

-# AI Generate
-
+# 视频自动翻译剪辑
 
 # 生成基础镜像，将所有依赖都放到基础镜像中
 <!-- v0 -->
-docker rmi harbor.5jstore.com:8020/ai/wm_generate_ai:v0
-docker build -f Dockerfile_v0 -t harbor.5jstore.com:8020/ai/wm_generate_ai:v0 .
+docker rmi harbor.5jstore.com:8020/ai/wm_video_cut:v0
+docker build -f Dockerfile_v0 -t harbor.5jstore.com:8020/ai/wm_video_cut:v0 .
 
 <!-- 0.1 根据v0手工生成，主要安装完整版的 cuda -->
-docker rmi harbor.5jstore.com:8020/ai/wm_generate_ai:v0.1
-docker run --gpus all --runtime=nvidia -it harbor.5jstore.com:8020/ai/wm_generate_ai:v0 /bin/bash
+docker rmi harbor.5jstore.com:8020/ai/wm_video_cut:v0.1
+docker run --gpus all --runtime=nvidia -it harbor.5jstore.com:8020/ai/wm_video_cut:v0 /bin/bash
 wget https://developer.download.nvidia.com/compute/cuda/11.3.0/local_installers/cuda_11.3.0_465.19.01_linux.run
 sh cuda_11.3.0_465.19.01_linux.run
 1、不选驱动
@@ -21,16 +20,16 @@ export CUDA_HOME=/usr/local/cuda-11.3
 删除下载文件
 rm cuda_11.3.0_465.19.01_linux.run
 提交新镜像
-docker commit c9c2a347491d harbor.5jstore.com:8020/ai/wm_generate_ai:v0.1
+docker commit c9c2a347491d harbor.5jstore.com:8020/ai/wm_video_cut:v0.1
 <!-- end -->
 
 <!-- 0.2 在0.1基础上安装项目依赖 -->
-docker rmi harbor.5jstore.com:8020/ai/wm_generate_ai:v0.2
-docker build -f Dockerfile_v0.2 -t harbor.5jstore.com:8020/ai/wm_generate_ai:v0.2 .
+docker rmi harbor.5jstore.com:8020/ai/wm_video_cut:v0.2
+docker build -f Dockerfile_v0.2 -t harbor.5jstore.com:8020/ai/wm_video_cut:v0.2 .
 
 # 生成主镜像
-docker rmi harbor.5jstore.com:8020/ai/wm_generate_ai:v2
-docker build -f Dockerfile -t harbor.5jstore.com:8020/ai/wm_generate_ai:v2 .
+docker rmi harbor.5jstore.com:8020/ai/wm_video_cut:v2
+docker build -f Dockerfile -t harbor.5jstore.com:8020/ai/wm_video_cut:v2 .
 
 docker-compose up -d
 
@@ -42,9 +41,9 @@ docker-compose up -d
 
 
 <!-- 调试docker镜像，去掉Dockerfile里最后一句启动命令 -->
-docker run -it harbor.5jstore.com:8020/ai/wm_generate_ai:v0 /bin/bash
-docker run -it harbor.5jstore.com:8020/ai/wm_generate_ai:v1 /bin/bash
-docker run -it harbor.5jstore.com:8020/ai/wm_generate_ai:v2 /bin/bash
+docker run -it harbor.5jstore.com:8020/ai/wm_video_cut:v0 /bin/bash
+docker run -it harbor.5jstore.com:8020/ai/wm_video_cut:v1 /bin/bash
+docker run -it harbor.5jstore.com:8020/ai/wm_video_cut:v2 /bin/bash
 
 docker run --gpus all --runtime=nvidia -v ./inputs/:/app/inputs/ -v ./outputs/:/app/outputs/ -it harbor.5jstore.com:8020/common/ai_generate_video:proc_v4 /bin/bash
 
@@ -52,4 +51,6 @@ ffmpeg -hwaccels
 
 <!-- 调试命令 -->
 python start.py
-gunicorn start:app -c ./gunicorn.conf.py
\ No newline at end of file
+gunicorn start:app -c ./gunicorn.conf.py
+
+flask --app start run --debug
\ No newline at end of file

app/utils/util.py

@@ -2,6 +2,7 @@
 # -*- coding: utf-8 -*-
 
 import requests
+from werkzeug.utils import secure_filename
 from urllib import parse
 import oss2
 import datetime, os
@@ -134,3 +135,23 @@ def get_wm_option(wm_key):
         return ""
     else:
         return wm_option.option_value
+
+
+def upload_file(file):
+    upload_dir = f'./inputs/{datetime.datetime.now().strftime("%Y%m%d%H%M%S")}'
+    if not os.path.exists(upload_dir):
+        os.makedirs(upload_dir)
+
+    def allowed_file(filename):
+        ALLOWED_EXTENSIONS = {'txt', 'pdf', 'png', 'jpg', 'jpeg', 'gif', 'mp4'}
+        return '.' in filename and \
+            filename.rsplit('.', 1)[1].lower() in ALLOWED_EXTENSIONS
+
+    if file and allowed_file(file.filename):
+        filename = secure_filename(file.filename)
+        filepath = os.path.join(upload_dir, filename)
+        file.save(filepath)
+        return filepath
+    else:
+        return ''
+

app/video_cut/autocut/utils.py

@@ -412,10 +412,12 @@ def text_to_audio_by_edge_tts(text, local_voice_file, lang, retry=1):
 
 def get_mp3_duration(file_path):
     from mutagen.mp3 import MP3
-
-    audio = MP3(file_path)
-    duration_in_seconds = audio.info.length
-    return duration_in_seconds
+    try:
+        audio = MP3(file_path)
+        duration_in_seconds = audio.info.length
+        return duration_in_seconds
+    except:
+        return 0
 
 
 def delete_files(current_dir, prefix):

app/video_cut/autocut/whisper_model.py

@@ -50,6 +50,7 @@ class WhisperModel(AbstractWhisperModel):
             "tiny", "base", "small", "medium", "large", "large-v2"
         ] = "small",
         device: Union[Literal["cpu", "cuda"], None] = None,
+        env = "dev",
     ):
         self.device = device
 

app/video_cut/autocut/wmdigit_cut.py

@@ -132,4 +132,5 @@ class Cutter:
         # utils.delete_files(os.path.dirname(fns['media']), f"._{os.path.splitext(os.path.basename(fns['media']))[0]}")
         # utils.delete_files(os.path.dirname(fns['media']), f".DS_Store")
         # utils.delete_files(os.path.dirname(fns['media']), f"._.DS_Store")
-        utils.delete_files(os.path.dirname(fns['media']), f".")
\ No newline at end of file
+        utils.delete_files(os.path.dirname(fns['media']), f".")
+        return output_fn
\ No newline at end of file

app/video_cut/autocut/wmdigit_transcribe.py

@@ -54,16 +54,15 @@ class Transcribe:
                 logging.info(f"Transcribed {input} to {output}")
                 self._save_md(name + ".md", output, input, bool(self.args.wmdigit))
                 logging.info(f'Saved texts to {name + ".md"} to mark sentences')
-            except:
-                if retry == 10:
-                    raise RuntimeError(f"Failed to Transcribing {input}")
+            except Exception as e:
+                if retry == 3:
+                    raise RuntimeError(f"Failed to Transcribing {e}")
                 else:
                     time.sleep(1)
                     logging.info(f"Retry {retry} to Transcribing {input}")
                     retry += 1
                     self.run(retry)
 
-
     def _detect_voice_activity(self, audio) -> List[SPEECH_ARRAY_INDEX]:
         """Detect segments that have voice activities"""
         if self.args.vad == "0":

app/video_cut/load_args.py

+import argparse
+import logging
+import os
+
+from app.video_cut.autocut import utils
+from app.video_cut.autocut.type import WhisperMode, WhisperModel
+
+
+def main_args(logger):
+    logger.info('load augument')
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--inputs", type=str, help="Inputs filenames/folders")
+    parser.add_argument(
+        "--lang",
+        type=str,
+        default="zh",
+        choices=[
+            "zh",
+            "en",
+            "Afrikaans",
+            "Arabic",
+            "Armenian",
+            "Azerbaijani",
+            "Belarusian",
+            "Bosnian",
+            "Bulgarian",
+            "Catalan",
+            "Croatian",
+            "Czech",
+            "Danish",
+            "Dutch",
+            "Estonian",
+            "Finnish",
+            "French",
+            "Galician",
+            "German",
+            "Greek",
+            "Hebrew",
+            "Hindi",
+            "Hungarian",
+            "Icelandic",
+            "Indonesian",
+            "Italian",
+            "Japanese",
+            "Kannada",
+            "Kazakh",
+            "Korean",
+            "Latvian",
+            "Lithuanian",
+            "Macedonian",
+            "Malay",
+            "Marathi",
+            "Maori",
+            "Nepali",
+            "Norwegian",
+            "Persian",
+            "Polish",
+            "Portuguese",
+            "Romanian",
+            "Russian",
+            "Serbian",
+            "Slovak",
+            "Slovenian",
+            "Spanish",
+            "Swahili",
+            "Swedish",
+            "Tagalog",
+            "Tamil",
+            "Thai",
+            "Turkish",
+            "Ukrainian",
+            "Urdu",
+            "Vietnamese",
+            "Welsh",
+        ],
+        help="The output language of transcription",
+    )
+    parser.add_argument(
+        "--prompt", type=str, default="", help="initial prompt feed into whisper"
+    )
+    parser.add_argument(
+        "--whisper-mode",
+        type=str,
+        default=WhisperMode.WHISPER.value,
+        choices=WhisperMode.get_values(),
+        help="Whisper inference mode: whisper: run whisper locally; openai: use openai api.",
+    )
+    parser.add_argument(
+        "--openai-rpm",
+        type=int,
+        default=3,
+        choices=[3, 50],
+        help="Openai Whisper API REQUESTS PER MINUTE(FREE USERS: 3RPM; PAID USERS: 50RPM). "
+        "More info: https://platform.openai.com/docs/guides/rate-limits/overview",
+    )
+    parser.add_argument(
+        "--whisper-model",
+        type=str,
+        default=WhisperModel.SMALL.value,
+        choices=WhisperModel.get_values(),
+        help="The whisper model used to transcribe.",
+    )
+    parser.add_argument(
+        "--bitrate",
+        type=str,
+        default="10m",
+        help="The bitrate to export the cutted video, such as 10m, 1m, or 500k",
+    )
+    parser.add_argument(
+        "--vad", help="If or not use VAD", choices=["1", "0", "auto"], default="0"
+    )
+    parser.add_argument(
+        "--force",
+        help="Force write even if files exist",
+        action=argparse.BooleanOptionalAction,
+    )
+    parser.add_argument(
+        "--encoding", type=str, default="utf-8", help="Document encoding format"
+    )
+    parser.add_argument(
+        "--device",
+        type=str,
+        default=None,
+        choices=["cpu", "cuda"],
+        help="Force to CPU or GPU for transcribing. In default automatically use GPU if available.",
+    )
+    parser.add_argument(
+        "--wmdigit",
+        help="Convert video to different language",
+        action=argparse.BooleanOptionalAction,
+    )
+
+    args = parser.parse_args(args=[])
+    args.wmdigit = True
+    args.force = True
+    args.vad = "0"
+
+    return args
\ No newline at end of file

app/video_cut/main.py

+from flask import abort, request, jsonify
+import datetime, os, time
+from app.video_cut.autocut import wmdigit_cut, wmdigit_transcribe, utils
+
+
+def error(msg):
+    abort(jsonify({"error": msg}))
+
+
+# 校验请求参数
+def validate_request():
+    if not request.json or 'instances' not in request.json:
+        error("参数错误: 缺少instances参数")
+
+    instance = request.json['instances']
+    if len(instance) <= 0 or 'video' not in instance[0] or 'lang' not in instance[0]:
+        error('参数错误: instances缺少:video,lang')
+
+    video = instance[0]['video']
+    lang = instance[0]['lang']
+
+    if len(video) <= 0:
+        error('参数错误: video 参数不可为空')
+    if len(lang) <= 0:
+        error('参数错误: lang 参数不可为空')
+
+    # if not video.startswith('http'):
+    #     error('video 必须是网络路径')
+
+    return video, lang
+
+
+# 主线
+def video_cut_pipeline(logger, args):
+    # print(args)
+    time_record = []
+    media_file, lang = validate_request()
+    all_start_time = time.time()
+
+    # 1、视频生成srt和md
+    start_time = time.time()
+    srt_fn = utils.change_ext(media_file, "srt")
+    md_fn = utils.change_ext(media_file, "md")
+    # 如果目标语言不是中文，则提示whisper翻译全部字幕
+    if lang != "zh":
+        prompt = f"Subtitles must be fully translated into {lang}"
+    else:
+        prompt = ""
+    logger.info(f"Transcribe {media_file} lang={lang} promt={prompt}")
+
+    args.inputs = [media_file]
+    args.lang = lang
+    wmdigit_transcribe.Transcribe(args).run()
+    time_record.append(f"视频生成srt和md。耗时: {time.time() - start_time:.4f} 秒")
+
+
+    # 2、从字幕生成cut视频
+    start_time = time.time()
+    args.inputs = [media_file, md_fn, srt_fn]
+    final_video_fn = wmdigit_cut.Cutter(args).run()
+    time_record.append(f"从字幕生成cut视频。耗时: {time.time() - start_time:.4f} 秒")
+    time_record.append(f"所有步骤处理完毕。耗时: {time.time() - all_start_time:.4f} 秒")
+    
+    for i in time_record:
+        print(i)
+
+    return final_video_fn, srt_fn
+
+
+

docker-compose-v1.yml

 version: '3'
 services:
   pytorch:    
-    image: "harbor.5jstore.com:8020/ai/wm_generate_ai:v1"
+    image: "harbor.5jstore.com:8020/ai/wm_video_cut:v1"
     restart: always
     runtime: nvidia
     environment:

docker-compose.yml

 version: '3'
 services:
   pytorch:    
-    image: "harbor.5jstore.com:8020/ai/wm_generate_ai:v2"
+    image: "harbor.5jstore.com:8020/ai/wm_video_cut:v2"
     restart: always
     runtime: nvidia
     environment:

gunicorn.conf.py

@@ -2,8 +2,4 @@ workers = 1    # 定义同时开启的处理请求的进程数量，根据网站
 worker_class = "gevent"   # 采用gevent库，支持异步处理请求，提高吞吐量
 timeout = 600
 
-bind = "0.0.0.0:5000"
-
-accesslog = '-'
-errorlog = '-'
-loglevel = 'debug'
\ No newline at end of file
+bind = "0.0.0.0:8080"

start.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
-from flask import Flask, abort, request, jsonify
+from flask import Flask, abort, request, jsonify, send_file
 import datetime, os, sys, time
 import logging
+from app.utils import util
 from app.models.wm_option import db
-
+from app.video_cut.load_args import main_args
+from app.video_cut.main import video_cut_pipeline
 
 logging.basicConfig(level = logging.INFO,format = '%(asctime)s - %(name)s - %(levelname)s - %(message)s')
 logger = logging.getLogger(__name__)
@@ -24,12 +26,37 @@ input_root = os.path.join(root, 'inputs')
 output_root = os.path.join(root, 'outputs')
 
 # 预加载模型
+args = main_args(logger)
 
 # 对外接口
-@app.route('/ai_generate_video', methods=['GET'])
-def ai_generate_video():
-    return jsonify({"result": "akakkakaka"})
+@app.route('/wm_video_cut', methods=['POST'])
+def wm_video_cut():
+    final_video_url, srt_url = video_cut_pipeline(logger, args)
+    return jsonify({"result": {"final_video_url": final_video_url, "srt_url": srt_url}})
+
+
+@app.route('/upload_file', methods=['POST'])
+def upload_file():
+    # check if the post request has the file part
+    if 'file' not in request.files:
+        return 'No file part', 403
+    file = request.files['file']
+    if file.filename == '':
+        return 'No selected file', 403
+    file_url = util.upload_file(file)
+    if not file_url:
+        return '上传文件失败', 500
+    else:
+        return jsonify({"result": {"file_url": file_url}})
+
+
+@app.route('/download/<path:file_path>')
+def download_file(file_path):
+    if not file_path.startswith('input'):
+        return '文件不存在', 404
+    # 发送文件给浏览器进行下载
+    return send_file(file_path, as_attachment=True)
+
 
 if __name__ == "__main__":
-    # 将host设置为0.0.0.0，则外网用户也可以访问到这个服务
-    app.run(debug=True, host="0.0.0.0", use_reloader=False)
+    app.run()