1. add progressbar \n 2.upload paddlespeech \n 3. add requirements.txt

PaddlePaddle_DeepSpeech2 @ 081f7807

+Subproject commit 081f7807a2ce0e12b98e6f0a0da0e650133f2d9e

detect_with_asr.py

-import shutil
 import time
 
 import openpyxl

narratage_detection.py

@@ -19,6 +19,7 @@ def trans_to_seconds(timepoint):
 
 
 def detect(video_path, start_time, end_time, book_path):
+    print("开始检测")
     if book_path is None:
         book_path = os.path.basename(video_path).split('.')[0] + ".xlsx"
     else:

requirements.txt

+azure_storage==0.37.0
+cn2an==0.5.16
+ffmpeg==1.4
+ffmpeg_python==0.2.0
+Flask==2.0.2
+flask_cors==3.0.10
+ijson==3.1.4
+LAC==2.1.2
+librosa==0.8.1
+mttkinter==0.6.1
+numpy==1.19.3
+opencv_python==4.5.4.58
+openpyxl==3.0.9
+paddle==1.0.2
+paddleocr==2.3.0.1
+paddlepaddle_gpu==2.2.0
+paddlespeech==0.1.0
+parakeet==0.24
+Pillow==9.0.0
+pyaudio==0.2.11
+pydub==0.25.1
+PyYAML==6.0
+resampy==0.2.2
+scipy==1.6.1
+SoundFile==0.10.3.post1
+tqdm==4.62.3
+visualdl==2.2.1
+webrtcvad==2.0.10
+yacs==0.1.8
+zhconv==1.4.3

speech_synthesis.py

 # coding=utf-8
 import os
 import argparse
+import time
 
 from azure.cognitiveservices.speech import AudioDataStream, SpeechConfig, SpeechSynthesizer
 from azure.cognitiveservices.speech.audio import AudioOutputConfig
@@ -44,10 +45,9 @@ def change_speed(wav_path, speed=1.5):
     :param speed: 转换后的语速
     :return:
     """
+    print("调速")
     cmd_line = 'ffmpeg -y -i {} -filter:a \"atempo={}\" {}'.format(tmp_file, speed, wav_path)
     os.system(cmd_line)
-    # 删除临时文件
-    os.remove(tmp_file)
 
 
 def read_sheet(book_path, sheet_name=None):
@@ -89,7 +89,7 @@ def get_narratage_text(sheet_content):
         if text is not None:
             if text == '翻译':
                 narratage_text.append(subtitle[i])
-                narratage_start_time.append(float(start_time[i]) + 0.1)
+                narratage_start_time.append(float(start_time[i]))
             else:
                 # 如果旁白中有换行符，即分为n段，则按照换行符进行分割，并间隔0.5s
                 text_split = text.split('\n')
@@ -97,7 +97,7 @@ def get_narratage_text(sheet_content):
                 for x in text_split:
                     narratage_text.append(x)
                     narratage_start_time.append(cur_start)
-                    cur_start = cur_start + len(x) / (4.5 * args.speed) + 0.5
+                    cur_start = cur_start + len(x) / 4.5 + 0.5
     return narratage_text, narratage_start_time
 
 
@@ -131,15 +131,20 @@ def export_caption(sheet_content, caption_file):
                 f.write(x + "\n\n")
 
 
-def ss_and_export():
+def ss_and_export(sheet_path, output_dir, speed, caption_file, state):
+    """
+    生成语音并导出字幕
+    :param sheet_path: 校对过的旁白脚本表格文件
+    :param output_dir: 存放音频文件的
+    :param speed:
+    :param caption_file:
+    :return:
+    """
+
     # 旁白解说表格的位置
-    book_path = args.sheet_path
+    book_path = sheet_path
     # 音频输出位置路径
-    root_path = args.output_dir
-    # 语速
-    speed = args.speed
-    # 字幕文件路径
-    caption_file = args.caption_file
+    root_path = output_dir
 
     # 如果文件夹不存在，则新建文件夹
     if not os.path.exists(root_path):
@@ -149,11 +154,19 @@ def ss_and_export():
     sheet_content = read_sheet(book_path)
     narratages, start_timepoint = get_narratage_text(sheet_content)
     export_caption(sheet_content, caption_file)
+    print("已导出旁白文件")
 
     # 生成旁白解说语音
     for i, text in enumerate(narratages):
         wav_path = os.path.join(root_path, '%.2f.wav' % start_timepoint[i])
         speech_synthesis(text, wav_path, speed)
+        print("目前正在处理{}".format(wav_path))
+        state[0] = float((i + 1) / len(narratages))
+
+    # 删除临时语音文件
+    if os.path.exists(tmp_file):
+        time.sleep(1)
+        os.remove(tmp_file)
 
 
 if __name__ == '__main__':
@@ -166,4 +179,4 @@ if __name__ == '__main__':
     args = parser.parse_args()
 
     # 主函数执行
-    ss_and_export(args.output_dir,args.sheet_path,args.caption_file,args.speed)
+    ss_and_export(args.output_dir, args.sheet_path, args.speed, args.caption_file)

split_wav.py

+# 最简单的是音轨分离,直接将背景音乐的轨道剥离,只剩下人声道后即可根据空白片段进行切割
+
+# 只有一个音轨时,使用音乐检索系统,分割人声和背景音乐声
+import os
+
+os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
+import numpy as np
+from moviepy.editor import *
+from pydub import AudioSegment
+from spleeter.audio.adapter import AudioAdapter
+from spleeter.separator import Separator
+
+audio_path = r'D:\AddCaption\disappearedBullet.wav'
+folder_name = r'D:\AddCaption\hysxm'
+split_len = 18
+
+
+def extract_audio(video_path, root, start_time, end_time):
+    """
+    从视频中读取音频
+    :param video_path: 视频存储路径
+    :param root: 存储临时文件的文件夹路径
+    :param start_time: 视频的实际开始时间
+    :param end_time: 视频的实际结束时间
+    :return: audio_path 提取的音频存储路径
+    """
+    global audio_path
+    audio_path = os.path.join(root, os.path.basename(video_path).split('.')[0] + ".wav")
+    if end_time == -1:
+        command = "ffmpeg -i {} -ar 16000 -ac 1 -ss {} -y {}".format(video_path, start_time, audio_path)
+    else:
+        command = "ffmpeg -i {} -ar 16000 -ac 1 -ss {} -to {} -y {}".format(video_path, start_time, end_time, audio_path)
+    os.system(command)
+    return audio_path
+
+
+# 将音频划分为n个文件，返回存储
+def split_audio():
+    parent = os.path.dirname(os.path.realpath(audio_path))
+    global folder_name
+    folder_name = os.path.join(parent, os.path.basename(audio_path)[:-4])
+    if not os.path.exists(folder_name):
+        os.mkdir(folder_name)
+    audio = AudioSegment.from_file(audio_path)
+    per_part = 300000
+    global split_len
+    split_len = int(np.ceil(len(audio) / per_part))
+    if audio.channels > 1:
+        audio, _ = audio.split_to_mono()
+    for i in range(1, split_len + 1):
+        cur_path = os.path.join(folder_name, '{}.wav'.format(str(i).zfill(2)))
+        audio[per_part * (i - 1):per_part * i].export(cur_path, format='wav')
+    return folder_name
+
+
+# 从音频中提取人声
+def extrac_speech():
+    separator = Separator('spleeter:2stems', multiprocess=False)
+    audio_adapter = AudioAdapter.default()
+
+    for i in range(1, split_len + 1):
+        audio_path = os.path.join(folder_name, '{}.wav'.format(str(i).zfill(2)))
+        out_path = os.path.join(folder_name, str(i).zfill(2))
+        if not os.path.exists(out_path):
+            os.mkdir(out_path)
+        waveform, framerate = audio_adapter.load(audio_path)
+        vocals = separator.separate(waveform)['vocals']
+        audio_adapter.save(os.path.join(out_path, 'vocal.wav'), vocals, framerate)
+
+
+# 提取背景音乐
+def extract_bgm():
+    mp4_path = 'test.wav'
+    snd = AudioSegment.from_file(mp4_path)
+    print(len(snd))
+    snd_l, snd_r = snd.split_to_mono()
+    print(len(snd_r), len(snd_l))
+    snd_r_inv = snd_r.invert_phase()  # 反相
+    bg_music2 = snd_r.overlay(snd_l.invert_phase())
+    bg_music = snd_l.overlay(snd_r_inv)  # 覆盖后获得背景音乐，但有噪音
+
+    print(len(bg_music))
+    bg_music.export('bgm.wav', format='wav')
+    total_music = snd_l.overlay(snd_r)
+    total_music.export("total.wav", format='wav')
+
+
+if __name__ == '__main__':
+    video_path = 'D:/heelo/shaolin.mkv'
+    extract_audio(video_path, './tmp', 62, 7489)
+    split_audio()
+    extrac_speech()