ViDove / src /task.py
Eason Lu
debug& add gradio web interface
d0f2803
raw
history blame
No virus
12.8 kB
import threading
import time
import openai
from pytube import YouTube
from os import getenv, getcwd
from pathlib import Path
from enum import Enum, auto
import logging
import subprocess
from src.srt_util.srt import SrtScript
from src.srt_util.srt2ass import srt2ass
from time import time, strftime, gmtime, sleep
from src.translators.translation import get_translation, prompt_selector
import torch
import stable_whisper
import shutil
"""
Youtube link
- link
- model
- output type
Video file
- path
- model
- output type
Audio file
- path
- model
- output type
"""
"""
TaskID
Progress: Enum
Computing resrouce status
SRT_Script : SrtScript
- input module -> initialize (ASR module)
- Pre-process
- Translation (%)
- Post process (time stamp)
- Output module: SRT_Script --> output(.srt)
- (Optional) mp4
"""
class TaskStatus(str, Enum):
CREATED = 'CREATED'
INITIALIZING_ASR = 'INITIALIZING_ASR'
PRE_PROCESSING = 'PRE_PROCESSING'
TRANSLATING = 'TRANSLATING'
POST_PROCESSING = 'POST_PROCESSING'
OUTPUT_MODULE = 'OUTPUT_MODULE'
class Task:
@property
def status(self):
with self.__status_lock:
return self.__status
@status.setter
def status(self, new_status):
with self.__status_lock:
self.__status = new_status
def __init__(self, task_id, task_local_dir, task_cfg):
self.__status_lock = threading.Lock()
self.__status = TaskStatus.CREATED
self.gpu_status = 0
openai.api_key = getenv("OPENAI_API_KEY")
self.task_id = task_id
self.task_local_dir = task_local_dir
self.ASR_setting = task_cfg["ASR"]
self.translation_setting = task_cfg["translation"]
self.translation_model = self.translation_setting["model"]
self.output_type = task_cfg["output_type"]
self.target_lang = task_cfg["target_lang"]
self.source_lang = task_cfg["source_lang"]
self.field = task_cfg["field"]
self.pre_setting = task_cfg["pre_process"]
self.post_setting = task_cfg["post_process"]
self.audio_path = None
self.SRT_Script = None
self.result = None
self.s_t = None
self.t_e = None
print(f"Task ID: {self.task_id}")
logging.info(f"Task ID: {self.task_id}")
logging.info(f"{self.source_lang} -> {self.target_lang} task in {self.field}")
logging.info(f"Translation Model: {self.translation_model}")
logging.info(f"subtitle_type: {self.output_type['subtitle']}")
logging.info(f"video_ouput: {self.output_type['video']}")
logging.info(f"bilingual_ouput: {self.output_type['bilingual']}")
logging.info("Pre-process setting:")
for key in self.pre_setting:
logging.info(f"{key}: {self.pre_setting[key]}")
logging.info("Post-process setting:")
for key in self.post_setting:
logging.info(f"{key}: {self.post_setting[key]}")
@staticmethod
def fromYoutubeLink(youtube_url, task_id, task_dir, task_cfg):
# convert to audio
logging.info("Task Creation method: Youtube Link")
return YoutubeTask(task_id, task_dir, task_cfg, youtube_url)
@staticmethod
def fromAudioFile(audio_path, task_id, task_dir, task_cfg):
# get audio path
logging.info("Task Creation method: Audio File")
return AudioTask(task_id, task_dir, task_cfg, audio_path)
@staticmethod
def fromVideoFile(video_path, task_id, task_dir, task_cfg):
# get audio path
logging.info("Task Creation method: Video File")
return VideoTask(task_id, task_dir, task_cfg, video_path)
# Module 1 ASR: audio --> SRT_script
def get_srt_class(self):
# Instead of using the script_en variable directly, we'll use script_input
# TODO: setup ASR module like translator
self.status = TaskStatus.INITIALIZING_ASR
self.t_s = time()
method = self.ASR_setting["whisper_config"]["method"]
whisper_model = self.ASR_setting["whisper_config"]["whisper_model"]
src_srt_path = self.task_local_dir.joinpath(f"task_{self.task_id}_{self.source_lang}.srt")
if not Path.exists(src_srt_path):
# extract script from audio
logging.info("extract script from audio")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
if method == "api":
with open(self.audio_path, 'rb') as audio_file:
transcript = openai.Audio.transcribe(model="whisper-1", file=audio_file, response_format="srt")
elif method == "stable":
model = stable_whisper.load_model(whisper_model, device)
transcript = model.transcribe(str(self.audio_path), regroup=False,
initial_prompt="Hello, welcome to my lecture. Are you good my friend?")
(
transcript
.split_by_punctuation(['.', '。', '?'])
.merge_by_gap(.15, max_words=3)
.merge_by_punctuation([' '])
.split_by_punctuation(['.', '。', '?'])
)
transcript = transcript.to_dict()
# after get the transcript, release the gpu resource
torch.cuda.empty_cache()
self.SRT_Script = SrtScript(self.source_lang, self.target_lang, transcript['segments'], self.field)
# save the srt script to local
self.SRT_Script.write_srt_file_src(src_srt_path)
# Module 2: SRT preprocess: perform preprocess steps
def preprocess(self):
self.status = TaskStatus.PRE_PROCESSING
logging.info("--------------------Start Preprocessing SRT class--------------------")
if self.pre_setting["sentence_form"]:
self.SRT_Script.form_whole_sentence()
if self.pre_setting["spell_check"]:
self.SRT_Script.spell_check_term()
if self.pre_setting["term_correct"]:
self.SRT_Script.correct_with_force_term()
processed_srt_path_src = str(Path(self.task_local_dir) / f'{self.task_id}_processed.srt')
self.SRT_Script.write_srt_file_src(processed_srt_path_src)
if self.output_type["subtitle"] == "ass":
logging.info("write English .srt file to .ass")
assSub_src = srt2ass(processed_srt_path_src, "default", "No", "Modest")
logging.info('ASS subtitle saved as: ' + assSub_src)
self.script_input = self.SRT_Script.get_source_only()
pass
def update_translation_progress(self, new_progress):
if self.progress == TaskStatus.TRANSLATING:
self.progress = TaskStatus.TRANSLATING.value[0], new_progress
# Module 3: perform srt translation
def translation(self):
logging.info("---------------------Start Translation--------------------")
prompt = prompt_selector(self.source_lang, self.target_lang, self.field)
get_translation(self.SRT_Script, self.translation_model, self.task_id, prompt, self.translation_setting['chunk_size'])
# Module 4: perform srt post process steps
def postprocess(self):
self.status = TaskStatus.POST_PROCESSING
logging.info("---------------------Start Post-processing SRT class---------------------")
if self.post_setting["check_len_and_split"]:
self.SRT_Script.check_len_and_split()
if self.post_setting["remove_trans_punctuation"]:
self.SRT_Script.remove_trans_punctuation()
logging.info("---------------------Post-processing SRT class finished---------------------")
# Module 5: output module
def output_render(self):
self.status = TaskStatus.OUTPUT_MODULE
video_out = self.output_type["video"]
subtitle_type = self.output_type["subtitle"]
is_bilingual = self.output_type["bilingual"]
results_dir =f"{self.task_local_dir}/results"
subtitle_path = f"{results_dir}/{self.task_id}_{self.target_lang}.srt"
self.SRT_Script.write_srt_file_translate(subtitle_path)
if is_bilingual:
subtitle_path = f"{results_dir}/{self.task_id}_{self.source_lang}_{self.target_lang}.srt"
self.SRT_Script.write_srt_file_bilingual(subtitle_path)
if subtitle_type == "ass":
logging.info("write .srt file to .ass")
subtitle_path = srt2ass(subtitle_path, "default", "No", "Modest")
logging.info('ASS subtitle saved as: ' + subtitle_path)
final_res = subtitle_path
# encode to .mp4 video file
if video_out and self.video_path is not None:
logging.info("encoding video file")
logging.info(f'ffmpeg comand: \nffmpeg -i {self.video_path} -vf "subtitles={subtitle_path}" {results_dir}/{self.task_id}.mp4')
subprocess.run(
["ffmpeg",
"-i", self.video_path,
"-vf", f"subtitles={subtitle_path}",
f"{results_dir}/{self.task_id}.mp4"])
final_res = f"{results_dir}/{self.task_id}.mp4"
self.t_e = time()
logging.info(
"Pipeline finished, time duration:{}".format(strftime("%H:%M:%S", gmtime(self.t_e - self.t_s))))
return final_res
def run_pipeline(self):
self.get_srt_class()
self.preprocess()
self.translation()
self.postprocess()
self.result = self.output_render()
# print(self.result)
class YoutubeTask(Task):
def __init__(self, task_id, task_local_dir, task_cfg, youtube_url):
super().__init__(task_id, task_local_dir, task_cfg)
self.youtube_url = youtube_url
def run(self):
yt = YouTube(self.youtube_url)
video = yt.streams.filter(progressive=True, file_extension='mp4').order_by('resolution').desc().first()
if video:
video.download(str(self.task_local_dir), filename=f"task_{self.task_id}.mp4")
logging.info(f'Video Name: {video.default_filename}')
else:
raise FileNotFoundError(f" Video stream not found for link {self.youtube_url}")
audio = yt.streams.filter(only_audio=True).first()
if audio:
audio.download(str(self.task_local_dir), filename=f"task_{self.task_id}.mp3")
else:
logging.info(" download audio failed, using ffmpeg to extract audio")
subprocess.run(
['ffmpeg', '-i', self.task_local_dir.joinpath(f"task_{self.task_id}.mp4"), '-f', 'mp3',
'-ab', '192000', '-vn', self.task_local_dir.joinpath(f"task_{self.task_id}.mp3")])
logging.info("audio extraction finished")
self.video_path = self.task_local_dir.joinpath(f"task_{self.task_id}.mp4")
self.audio_path = self.task_local_dir.joinpath(f"task_{self.task_id}.mp3")
logging.info(f" Video File Dir: {self.video_path}")
logging.info(f" Audio File Dir: {self.audio_path}")
logging.info(" Data Prep Complete. Start pipeline")
super().run_pipeline()
class AudioTask(Task):
def __init__(self, task_id, task_local_dir, task_cfg, audio_path):
super().__init__(task_id, task_local_dir, task_cfg)
# TODO: check audio format
self.audio_path = audio_path
self.video_path = None
def run(self):
logging.info(f"Video File Dir: {self.video_path}")
logging.info(f"Audio File Dir: {self.audio_path}")
logging.info("Data Prep Complete. Start pipeline")
super().run_pipeline()
class VideoTask(Task):
def __init__(self, task_id, task_local_dir, task_cfg, video_path):
super().__init__(task_id, task_local_dir, task_cfg)
# TODO: check video format {.mp4}
new_video_path = f"{task_local_dir}/task_{self.task_id}.mp4"
print(new_video_path)
logging.info(f"Copy video file to: {new_video_path}")
shutil.copyfile(video_path, new_video_path)
self.video_path = new_video_path
def run(self):
logging.info("using ffmpeg to extract audio")
subprocess.run(
['ffmpeg', '-i', self.video_path, '-f', 'mp3',
'-ab', '192000', '-vn', self.task_local_dir.joinpath(f"task_{self.task_id}.mp3")])
logging.info("audio extraction finished")
self.audio_path = self.task_local_dir.joinpath(f"task_{self.task_id}.mp3")
logging.info(f" Video File Dir: {self.video_path}")
logging.info(f" Audio File Dir: {self.audio_path}")
logging.info("Data Prep Complete. Start pipeline")
super().run_pipeline()