from openai import AzureOpenAI
from langchain_openai import AzureChatOpenAI
from huggingface_hub import InferenceClient
import os
import ffmpeg
from typing import List
from moviepy.editor import VideoFileClip
import nltk
from sklearn.feature_extraction.text import TfidfVectorizer
from langchain import HuggingFaceHub, PromptTemplate, LLMChain
import gradio as gr
from pytube import YouTube
import requests
import logging
import os
nltk.download('punkt')
nltk.download('stopwords')
class VideoAnalytics:
"""
Class for performing analytics on videos including transcription, summarization, topic generation,
and extraction of important sentences.
"""
def __init__(self):
"""
Initialize the VideoAnalytics object.
Args:
hf_token (str): Hugging Face API token.
"""
# Initialize AzureOpenAI client
self.client = AzureOpenAI()
self.mistral_client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
# Initialize transcribed text variable
self.transcribed_text = ""
# API URL for accessing the Hugging Face model
self.API_URL = "https://api-inference.huggingface.co/models/openai/whisper-large-v3"
hf_token = os.getenv('HF_TOKEN')
# Placeholder for Hugging Face API token
self.hf_token = hf_token # Replace this with the actual Hugging Face API token
# Set headers for API requests with Hugging Face token
self.headers = {"Authorization": f"Bearer {self.hf_token}"}
# Initialize english text variable
self.english_text = ""
self.openai_llm = AzureChatOpenAI(
deployment_name="ChatGPT",
)
# Configure logging settings
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
def transcribe_video(self, vid: str) -> str:
"""
Transcribe the audio of the video.
Args:
vid (str): Path to the video file.
Returns:
str: Transcribed text.
"""
try:
# Load the video file and extract audio
video = VideoFileClip(vid)
audio = video.audio
# Write audio to a temporary file
audio.write_audiofile("output_audio.mp3")
audio_file = open("output_audio.mp3", "rb")
# Define a helper function to query the Hugging Face model
def query(data):
response = requests.post(self.API_URL, headers=self.headers, data=data)
return response.json()
# Send audio data to the Hugging Face model for transcription
output = query(audio_file)
print(output)
# Update the transcribed_text attribute with the transcription result
self.transcribed_text = output["text"]
# Update the translation text into english_text
self.english_text = self.translation()
# Return the transcribed text
return output["text"]
except Exception as e:
logging.error(f"Error transcribing video: {e}")
return ""
def generate_video_summary(self) -> str:
"""
Generate a summary of the transcribed video.
Returns:
str: Generated summary.
"""
try:
# Define a conversation between system and user
conversation = [
{"role": "system", "content": "You are a Summarizer"},
{"role": "user", "content": f"""summarize the following text delimited by triple backticks.Output must in english.
In two format of Outputs given below:
Abstractive Summary:
Extractive Summary:
```{self.english_text}```
"""}
]
# Generate completion using ChatGPT model
response = self.client.chat.completions.create(
model="ChatGPT",
messages=conversation,
temperature=0,
max_tokens=1000
)
# Get the generated summary message
message = response.choices[0].message.content
return message
except Exception as e:
logging.error(f"Error generating video summary: {e}")
return ""
def generate_topics(self) -> str:
"""
Generate topics from the transcribed video.
Returns:
str: Generated topics.
"""
try:
# Define a conversation between system and user
conversation = [
{"role": "system", "content": "You are a Topic Generator"},
{"role": "user", "content": f"""generate single Topics from the following text don't make sentence for topic generation,delimited by triple backticks.Output must in english.
list out the topics:
Topics:
```{self.english_text}```
"""}
]
# Generate completion using ChatGPT model
response = self.client.chat.completions.create(
model="ChatGPT",
messages=conversation,
temperature=0,
max_tokens=1000
)
# Get the generated topics message
message = response.choices[0].message.content
return message
except Exception as e:
logging.error(f"Error generating topics: {e}")
return ""
def translation(self) -> str:
"""
translation from the transcribed video.
Returns:
str: translation.
"""
try:
# Define a conversation between system and user
conversation = [
{"role": "system", "content": "You are a Multilingual Translator"},
{"role": "user", "content": f""" Translate the following text in English ,delimited by triple backticks.
```{self.transcribed_text}```
"""}
]
# Generate completion using ChatGPT model
response = self.client.chat.completions.create(
model="ChatGPT",
messages=conversation,
temperature=0,
max_tokens=1000
)
# Get the generated topics message
message = response.choices[0].message.content
return message
except Exception as e:
logging.error(f"Error generating topics: {e}")
return ""
def format_prompt(self, question: str, data: str) -> str:
"""
Formats the prompt for the language model.
Args:
question (str): The user's question.
data (str): The data to be analyzed.
Returns:
str: Formatted prompt.
"""
prompt = ""
prompt = f"""[INST] you are the german language and universal language expert .your task is analyze the given data and user ask any question about given data answer to the user question.your returning answer must in user's language.otherwise reply i don't know.
data:{data}
question:{question}[/INST]"""
prompt1 = f"[INST] {question} [/INST]"
return prompt+prompt1
def generate(self, prompt: str, transcribed_text: str, temperature=0.9, max_new_tokens=5000, top_p=0.95,
repetition_penalty=1.0) -> str:
"""
Generates text based on the prompt and transcribed text.
Args:
prompt (str): The prompt for generating text.
transcribed_text (str): The transcribed text for analysis.
temperature (float): Controls the randomness of the sampling. Default is 0.9.
max_new_tokens (int): Maximum number of tokens to generate. Default is 5000.
top_p (float): Nucleus sampling parameter. Default is 0.95.
repetition_penalty (float): Penalty for repeating the same token. Default is 1.0.
Returns:
str: Generated text.
"""
try:
temperature = float(temperature)
if temperature < 1e-2:
temperature = 1e-2
top_p = float(top_p)
generate_kwargs = dict(
temperature=temperature,
max_new_tokens=max_new_tokens,
top_p=top_p,
repetition_penalty=repetition_penalty,
do_sample=True,
seed=42,
)
# Format the prompt
formatted_prompt = self.format_prompt(prompt,transcribed_text)
# Generate text using the mistral client
stream = self.mistral_client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
output = ""
# Concatenate generated text
for response in stream:
output += response.token.text
return output.replace("","")
except Exception as e:
logging.error(f"Error in text generation: {e}")
return "An error occurred during text generation."
def video_qa(self, question: str, model: str) -> str:
"""
Performs video question answering.
Args:
question (str): The question asked by the user.
model (str): The language model to be used ("OpenAI" or "Mixtral").
Returns:
str: Answer to the user's question.
"""
try:
if model == "OpenAI":
template = """you are the universal language expert .your task is analyze the given text and user ask any question about given text answer to the user question.otherwise reply i don't know.
extracted_text:{text}
user_question:{question}"""
prompt = PromptTemplate(template=template, input_variables=["text","question"])
llm_chain = LLMChain(prompt=prompt, verbose=True, llm=self.openai_llm)
# Run the language model chain
result = llm_chain.run({"text":self.english_text,"question":question})
return result
elif model == "Mixtral":
# Generate answer using Mixtral model
result = self.generate(question,self.english_text)
return result
except Exception as e:
logging.error(f"Error in video question answering: {e}")
return "An error occurred during video question answering."
def extract_video_important_sentence(self) -> str:
"""
Extract important sentences from the transcribed video.
Returns:
str: Extracted important sentences.
"""
try:
# Tokenize the sentences
sentences = nltk.sent_tokenize(self.english_text)
# Initialize TF-IDF vectorizer
tfidf_vectorizer = TfidfVectorizer()
# Fit the vectorizer on the summary sentences
tfidf_matrix = tfidf_vectorizer.fit_transform(sentences)
# Calculate sentence scores based on TF-IDF values
sentence_scores = tfidf_matrix.sum(axis=1)
# Create a list of (score, sentence) tuples
sentence_rankings = [(score, sentence) for score, sentence in zip(sentence_scores, sentences)]
# Sort sentences by score in descending order
sentence_rankings.sort(reverse=True)
# Set a threshold for selecting sentences
threshold = 2.5 # Adjust as needed
# Select sentences with scores above the threshold
selected_sentences = [sentence for score, sentence in sentence_rankings if score >= threshold]
# Join selected sentences to form the summary
summary = '\n\n'.join(selected_sentences)
return summary
except Exception as e:
logging.error(f"Error extracting important sentences: {e}")
return ""
def write_text_files(self, text: str, filename: str) -> None:
"""
Write text to a file.
Args:
text (str): Text to be written to the file.
filename (str): Name of the file.
"""
try:
file_path = f"{filename}.txt"
with open(file_path, 'w') as file:
# Write content to the file
file.write(text)
except Exception as e:
logging.error(f"Error writing text to file: {e}")
def Download(self, link: str) -> str:
"""
Download a video from YouTube.
Args:
link (str): YouTube video link.
Returns:
str: Path to the downloaded video file.
"""
try:
# Initialize YouTube object with the provided link
youtubeObject = YouTube(link)
# Get the highest resolution stream
youtubeObject = youtubeObject.streams.get_highest_resolution()
try:
# Attempt to download the video
file_name = youtubeObject.download()
return file_name
except:
# Log any errors that occur during video download
logging.info("An error has occurred")
logging.info("Download is completed successfully")
except Exception as e:
# Log any errors that occur during initialization of YouTube object
logging.error(f"Error downloading video: {e}")
return ""
def save_audio_with_gtts(self,text, filename):
tts = gTTS(text=text, lang='en')
tts.save(filename)
return filename
def main(self, video: str = None, input_path: str = None) -> tuple:
"""
Perform video analytics.
Args:
video (str): Path to the video file.
input_path (str): Input path for the video.
Returns:
tuple: Summary, important sentences, and topics.
"""
try:
# Download the video if input_path is provided, otherwise use the provided video path
if input_path:
input_path = self.Download(input_path)
video_ = VideoFileClip(input_path)
duration = video_.duration
video_.close()
if round(duration) <= 600:
text = self.transcribe_video(input_path)
else:
return "Video Duration Above 10 Minutes,Try Below 10 Minutes Video","",""
elif video:
video_ = VideoFileClip(video)
duration = video_.duration
video_.close()
if round(duration) <= 600:
text = self.transcribe_video(video)
input_path = video
else:
return "Video Duration Above 10 Minutes,Try Below 10 Minutes Video","",""
# Generate summary, important sentences, and topics
summary = self.generate_video_summary()
self.write_text_files(summary,"Summary")
summary_voice = save_audio_with_gtts(summary,"summary.mp3")
important_sentences = self.extract_video_important_sentence()
self.write_text_files(important_sentences,"Important_Sentence")
important_sentences_voice = save_audio_with_gtts(important_sentences,"important_sentences.mp3")
topics = self.generate_topics()
self.write_text_files(topics,"Topics")
topics_voice = save_audio_with_gtts(topics,"topics.mp3")
# Return the generated summary, important sentences, and topics
return summary,important_sentences,topics,summary_voice,important_sentences_voice,topics_voice
except Exception as e:
# Log any errors that occur during video analytics
logging.error(f"Error in main function: {e}")
return "", "", ""
def gradio_interface(self):
with gr.Blocks(css="style.css",theme=gr.themes.Soft()) as demo:
gr.HTML("""Video Analytics
""")
with gr.Row():
yt_link = gr.Textbox(label= "Youtube Link",placeholder="https://www.youtube.com/watch?v=")
with gr.Row():
video = gr.Video(sources="upload",height=200,width=300)
with gr.Row():
submit_btn = gr.Button(value="Submit")
with gr.Tab("Summary"):
with gr.Row():
summary = gr.Textbox(show_label=False,lines=10)
with gr.Row():
summary_download = gr.DownloadButton(label="Download",value="Summary.txt",visible=True,size='lg',elem_classes="download_button")
with gr.Row():
summary_audio = gr.Audio(show_label= False,elem_classes='audio_class')
with gr.Tab("Important Sentences"):
with gr.Row():
Important_Sentences = gr.Textbox(show_label=False,lines=10)
with gr.Row():
sentence_download = gr.DownloadButton(label="Download",value="Important_Sentence.txt",visible=True,size='lg',elem_classes="download_button")
with gr.Row():
important_sentence_audio = gr.Audio(show_label = False,elem_classes='audio_class')
with gr.Tab("Topics"):
with gr.Row():
Topics = gr.Textbox(show_label=False,lines=10)
with gr.Row():
topics_download = gr.DownloadButton(label="Download",value="Topics.txt",visible=True,size='lg',elem_classes="download_button")
with gr.Row():
topics_audio = gr.Audio(show_label=False,elem_classes='audio_class')
with gr.Tab("Video QA"):
with gr.Row():
with gr.Column(scale=0.70):
question = gr.Textbox(show_label=False,placeholder="Ask Your Questions...")
with gr.Column(scale=0.30):
model = gr.Dropdown(["OpenAI", "Mixtral"],show_label=False,value="model")
with gr.Row():
result = gr.Textbox(label='Answer',lines=10)
submit_btn.click(self.main,[video,yt_link],[summary,Important_Sentences,Topics,summary_audio,important_sentence_audio,topics_audio])
question.submit(self.video_qa,[question,model],result)
demo.launch()
if __name__ == "__main__":
video_analytics = VideoAnalytics()
video_analytics.gradio_interface()