Update functions.py

functions.py

@@ -1,6 +1,7 @@
 import whisper
 import os
 from pytube import YouTube
+import openai
 import pandas as pd
 import plotly_express as px
 import nltk
@@ -21,6 +22,7 @@ import pickle, math
 import wikipedia
 from pyvis.network import Network
 import torch
+from pydub import AudioSegment
 from langchain.docstore.document import Document
 from langchain.embeddings import HuggingFaceEmbeddings,HuggingFaceInstructEmbeddings
 from langchain.vectorstores import FAISS
@@ -218,14 +220,42 @@ def inference(link, upload, _asr_model):
     
     if validators.url(link):
     
-      yt = YouTube(link)
-      title = yt.title
-      audio_file = yt.streams.filter(only_audio=True)[0].download(filename="audio.mp4")
+        yt = YouTube(link)
+        title = yt.title
 
-      file_size = os.path.getsize(path)
+        #Get audio file from YT
+        audio_file = yt.streams.filter(only_audio=True)[0].download(filename="audio.mp4")
 
-      if file_size > 25:
-          nnn
+        #Get size of audio file
+        audio_size = round(os.path.getsize(path)/(1024*1024),1)
+
+        #Check if file is > 24mb, if not then use Whisper API
+        if audio_size <= 24:
+
+            #Use whisper API
+            transcript = openai.Audio.translate("whisper-1", audio_file)
+
+        else:
+
+            st.write('File size larger than 24mb, applying chunking and transcription')
+
+            # load the audio file
+            audio_file = AudioSegment.from_file(path, format="mp4")
+            
+            # set chunk size to 24mb (in bytes)
+            chunk_size = 24 * 1024 * 1024
+
+            # create a directory to store the output files
+            if not os.path.exists("audio_output"):
+                os.mkdir("audio_output")
+
+            audio_chunks = []
+            
+            # iterate over each chunk and export it as a separate file
+            for i, chunk in enumerate(audio_file[::chunk_size]):
+                chunk.export(f"output/chunk_{i}.mp4", format="mp4")
+                audio_chunks.append(f"output/chunk_{i}.mp4")
+                
 
       results = _asr_model.transcribe(path, task='transcribe', language='en')