diff --git a/m3u8.py b/m3u8.py deleted file mode 100644 index aeb007e..0000000 --- a/m3u8.py +++ /dev/null @@ -1,323 +0,0 @@ -# 4.08.2023 -> 14.09.2023 -> 17.09.2023 -> 3.12.2023 - -# Import -import re, os, sys, glob, time, requests, shutil, ffmpeg, subprocess -from functools import partial -from multiprocessing.dummy import Pool -from tqdm.rich import tqdm -import moviepy.editor as mp -from Stream.util.platform import * - -# Class import -#from Stream.util.console import console -from Stream.util.console import console - -# Disable warning -import warnings -from tqdm import TqdmExperimentalWarning -warnings.filterwarnings("ignore", category=TqdmExperimentalWarning) - -# Variable -main_out_folder = "videos" -os.makedirs("videos", exist_ok=True) - -# [ decoder ] -> costant = ou_ts -class Video_Decoder(object): - - iv = "" - uri = "" - method = "" - - def __init__(self, x_key, uri): - self.method = x_key["METHOD"] if "METHOD" in x_key.keys() else "" - self.uri = uri - self.iv = x_key["IV"].lstrip("0x") if "IV" in x_key.keys() else "" - - def decode_aes_128(self, video_fname: str): - if is_platform_linux(): - frame_name = video_fname.split("/")[-1].split("-")[0] + ".ts" - else: - frame_name = video_fname.split("\\")[-1].split("-")[0] + ".ts" - res_cmd = subprocess.run(["openssl","aes-128-cbc","-d","-in", video_fname,"-out", "ou_ts/"+frame_name,"-nosalt","-iv", self.iv,"-K", self.uri ], capture_output=True) - - res_cmd_str = res_cmd.stderr.decode("utf-8") - res_cmd_fix = res_cmd_str.replace("b", "").replace("\n", "").replace("\r", "") - - if "lengthad" in res_cmd_fix: - console.log("[red]Wrong m3u8 key or remove key from input !!!") - sys.exit(0) - -def decode_ext_x_key(key_str: str): - key_str = key_str.replace('"', '').lstrip("#EXT-X-KEY:") - v_list = re.findall(r"[^,=]+", key_str) - key_map = {v_list[i]: v_list[i+1] for i in range(0, len(v_list), 2)} - return key_map - - -# [ util ] -def save_in_part(folder_ts, merged_mp4, file_extension = ".ts"): - - # Get list of ts file in order - os.chdir(folder_ts) - - - # Order all ts file - try: ordered_ts_names = sorted(glob.glob(f"*{file_extension}"), key=lambda x:float(re.findall("(\d+)", x.split("_")[1])[0])) - except: - try: ordered_ts_names = sorted(glob.glob(f"*{file_extension}"), key=lambda x:float(re.findall("(\d+)", x.split("-")[1])[0])) - except: ordered_ts_names = sorted(glob.glob(f"*{file_extension}")) - - open("concat.txt", "wb") - open("part_list.txt", "wb") - - # Variable for download - list_mp4_part = [] - part = 0 - start = 0 - end = 200 - - # Create mp4 from start ts to end - def save_part_ts(start, end, part): - #console.log(f"[blue]Process part [green][[red]{part}[green]]") - list_mp4_part.append(f"{part}.mp4") - - - with open(f"{part}_concat.txt", "w") as f: - for i in range(start, end): - f.write(f"file {ordered_ts_names[i]} \n") - # try: - # ffmpeg.input(f"{part}_concat.txt", format='concat', safe=0).output(f"{part}.mp4", c='copy', loglevel="quiet").run(capture_stdout=True, capture_stderr=True) - # except ffmpeg.Error as e: - # print('stderr:', e.stderr.decode('utf8')) - ffmpeg.input(f"{part}_concat.txt", format='concat', safe=0).output(f"{part}.mp4", c='copy', loglevel="quiet").run(capture_stdout=True, capture_stderr=True) - - - # Save first part - save_part_ts(start, end, part) - - # Save all other part - for _ in range(start, end): - - # Increment progress ts file - start+= 200 - end += 200 - part+=1 - - # Check if end or not - if(end < len(ordered_ts_names)): - save_part_ts(start, end, part) - else: - save_part_ts(start, len(ordered_ts_names), part) - break - - # Merge all part - console.log(f"[purple]Merge all: {file_extension} file") - with open("part_list.txt", 'w') as f: - for mp4_fname in list_mp4_part: - f.write(f"file {mp4_fname}\n") - - ffmpeg.input("part_list.txt", format='concat', safe=0).output(merged_mp4, c='copy', loglevel="quiet").run() - - # try: - # ffmpeg.input("part_list.txt", format='concat', safe=0).output(merged_mp4, c='copy', loglevel="quiet").run(capture_stdout=True, capture_stderr=True) - # except ffmpeg.Error as e: - # print('stderr:', e.stderr.decode('utf8')) - -def download_ts_file(ts_url: str, store_dir: str, headers): - - # Get ts name and folder - ts_name = ts_url.split('/')[-1].split("?")[0] - ts_dir = store_dir + "/" + ts_name - - # Check if exist - if(not os.path.isfile(ts_dir)): - - # Download - ts_res = requests.get(ts_url, headers=headers) - - if(ts_res.status_code == 200): - with open(ts_dir, 'wb+') as f: - f.write(ts_res.content) - else: - print(f"Failed to download streaming file: {ts_name}.") - - time.sleep(0.5) - -def download_vvt_sub(content, language, folder_id): - - # Get content of vvt - url_main_sub = "" - vvt = content.split("\n") - - # Find main url or vvt - for i in range(len(vvt)): - line = str(vvt[i]) - - if line.startswith("#EXTINF"): - url_main_sub = vvt[i+1] - - # Save subtitle to main folder out - path = os.path.join(main_out_folder, str(folder_id)) - os.makedirs(path, exist_ok=True) - open(os.path.join(path, "sub_"+str(language)+".vtt"), "wb").write(requests.get(url_main_sub).content) - -# [ donwload ] -def dw_m3u8(m3u8_link, m3u8_content, m3u8_headers="", decrypt_key="", merged_mp4="test.mp4"): - - # Reading the m3u8 file - m3u8_http_base = m3u8_link.rstrip(m3u8_link.split("/")[-1]) - m3u8 = m3u8_content.split('\n') - ts_url_list = [] - ts_names = [] - x_key_dict = dict() - is_encryped = False - - # Parsing the content in m3u8 with creation of url_list with url of ts file - for i_str in range(len(m3u8)): - line_str = m3u8[i_str] - - if "AES-128" in str(line_str): - is_encryped = True - - if line_str.startswith("#EXT-X-KEY:"): - x_key_dict = decode_ext_x_key(line_str) - - if line_str.startswith("#EXTINF"): - ts_url = m3u8[i_str+1] - ts_names.append(ts_url.split('/')[-1]) - - if not ts_url.startswith("http"): - ts_url = m3u8_http_base + ts_url - - ts_url_list.append(ts_url) - #console.log(f"[blue]Find [white]=> [red]{len(ts_url_list)}[blue] ts file to download") - #console.log(f"[green]Is m3u8 encryped => [red]{is_encryped}") - - if is_encryped and decrypt_key == "": - console.log(f"[red]M3U8 Is encryped") - sys.exit(0) - - if is_encryped: - #console.log(f"[blue]Use decrypting") - - video_decoder = Video_Decoder(x_key=x_key_dict, uri=decrypt_key) - os.makedirs("ou_ts", exist_ok=True) - - # Using multithreading to download all ts file - os.makedirs("temp_ts", exist_ok=True) - pool = Pool(15) - gen = pool.imap(partial(download_ts_file, store_dir="temp_ts", headers=m3u8_headers), ts_url_list) - for _ in tqdm(gen, total=len(ts_url_list), unit="bytes", unit_scale=True, unit_divisor=1024, desc="[yellow]Download"): - pass - pool.close() - pool.join() - - if is_encryped: - for ts_fname in tqdm(glob.glob("temp_ts/*.ts"), desc="[yellow]Decoding"): - video_decoder.decode_aes_128(ts_fname) - - # Start to merge all *.ts files - save_in_part("ou_ts", merged_mp4) - else: - save_in_part("temp_ts", merged_mp4) - - - # Clean temp file - os.chdir("..") - console.log("[green]Clean") - - - if is_platform_linux(): - if is_encryped: - shutil.move("ou_ts/"+merged_mp4 , main_out_folder+"/") - else: - shutil.move("temp_ts/"+merged_mp4 , main_out_folder+"/") - - else: - if is_encryped: - shutil.move("ou_ts\\"+merged_mp4 , main_out_folder+"\\") - else: - shutil.move("temp_ts\\"+merged_mp4 , main_out_folder+"\\") - - shutil.rmtree("ou_ts", ignore_errors=True) - shutil.rmtree("temp_ts", ignore_errors=True) - -def dw_aac(m3u8_link, m3u8_content, m3u8_headers, merged_mp3): - - # Reading the m3u8 file - url_base = m3u8_link.rstrip(m3u8_link.split("/")[-1]) - m3u8 = m3u8_content.split('\n') - ts_url_list = [] - ts_names = [] - - # Parsing the content in m3u8 with creation of url_list with url of ts file - for i in range(len(m3u8)): - line = m3u8[i] - - if line.startswith("#EXTINF"): - ts_url = m3u8[i+1] - ts_names.append(ts_url.split('/')[-1]) - - if not ts_url.startswith("http"): - ts_url = url_base + ts_url - - ts_url_list.append(ts_url) - console.log(f"[blue]Find [white]=> [red]{len(ts_url_list)}[blue] ts file to download") - - # Using multithreading to download all ts file - os.makedirs("temp_ts", exist_ok=True) - pool = Pool(15) - gen = pool.imap(partial(download_ts_file, store_dir="temp_ts", headers=m3u8_headers), ts_url_list) - for _ in tqdm(gen, total=len(ts_url_list), unit="bytes", unit_scale=True, unit_divisor=1024, desc="[yellow]Download"): - pass - pool.close() - pool.join() - - save_in_part("temp_ts", merged_mp3, file_extension=".aac") - - # Clean temp file - os.chdir("..") - console.log("[green]Clean") - shutil.move("temp_ts\\"+merged_mp3 , ".") - - shutil.rmtree("ou_ts", ignore_errors=True) - shutil.rmtree("temp_ts", ignore_errors=True) - -def dw_vvt_sub(url, headers, folder_id) -> (None): - - print(url, headers, folder_id) - - # Get content of m3u8 vvt - req = requests.get(url, headers=headers) - vvts = req.text.split('\n') - vvt_data = [] - - # Parsing the content in m3u8 of vvt with creation of url_list with url and name of language - for line in vvts: - line = line.split(",") - if line[0] == "#EXT-X-MEDIA:TYPE=SUBTITLES": - - vvt_data.append({ - 'language': line[2].split("=")[1].replace('"', ""), - 'url': line[-1].split("URI=")[1].replace('"', "") - }) - - - # Check array is not empty - if len(vvt_data) > 0: - - # Download all subtitle - for i in range(len(vvts)): - console.log(f"[blue]Download [red]sub => [green]{vvt_data[i]['language']}") - download_vvt_sub(requests.get(vvt_data[i]['url']).text, vvt_data[i]['language'], folder_id) - - else: - console.log("[red]Cant find info of subtitle [SKIP]") - -def join_audio_to_video(audio_path, video_path, out_path): - - audio = mp.AudioFileClip(audio_path) - video1 = mp.VideoFileClip(video_path) - final = video1.set_audio(audio) - - final.write_videofile(out_path) \ No newline at end of file