Spaces:
Build error
Build error
app.py
CHANGED
@@ -1,13 +1,97 @@
|
|
1 |
import gradio as gr
|
2 |
-
from
|
3 |
-
from
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
#https://www.youtube.com/watch?v=smUHQndcmOY&t=425s
|
7 |
#html = HTML("<iframe width='560' height='315' src='https://www.youtube.com/watch?v=smUHQndcmOY&t=425s' frameborder='0' allowfullscreen></iframe>")
|
8 |
#html = "<iframe width='560' height='315' src='https://www.youtube.com/embed/smUHQndcmOY' frameborder='0' allowfullscreen></iframe>"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
#vid = YouTubeVideo('smUHQndcmOY&t=425s')
|
10 |
return html
|
|
|
11 |
#https://youtu.be/smUHQndcmOY
|
12 |
def fun(url):
|
13 |
return gr.Video(value=url)
|
@@ -24,13 +108,13 @@ with demo:
|
|
24 |
)
|
25 |
with gr.Row():
|
26 |
input_url = gr.Textbox() #gr.HTML(placeholder="Enter a video link here..")
|
27 |
-
|
28 |
output_vid = gr.HTML()
|
29 |
|
30 |
b1 = gr.Button("Publish Video")
|
31 |
#b2 = gr.Button("Generate Image")
|
32 |
|
33 |
-
b1.click(display_vid, input_url, output_vid)
|
34 |
#b2.click(poem_to_image, poem_txt, output_image)
|
35 |
#examples=examples
|
36 |
|
|
|
1 |
import gradio as gr
|
2 |
+
from youtube_transcript_api import YouTubeTranscriptApi
|
3 |
+
from transformers import AutoTokenizer
|
4 |
+
from transformers import pipeline
|
5 |
+
from transformers import AutoModelForQuestionAnswering
|
6 |
+
import pandas as pd
|
7 |
+
from sentence_transformers import SentenceTransformer, util
|
8 |
+
import torch
|
9 |
|
10 |
+
#from IPython.display import HTML, IFrame
|
11 |
+
#from IPython.display import YouTubeVideo
|
12 |
+
|
13 |
+
#input - video link, output - full transcript
|
14 |
+
def get_transcript(link):
|
15 |
+
video_id = link.split("=")[1]
|
16 |
+
print(f"video id extracted is : {video_id}")
|
17 |
+
transcript = YouTubeTranscriptApi.get_transcript(video_id)
|
18 |
+
FinalTranscript = ' '.join([i['text'] for i in transcript])
|
19 |
+
return transcript, video_id
|
20 |
+
|
21 |
+
|
22 |
+
#input - question and transcript, output - answer timestamp
|
23 |
+
def get_answers_timestamp(question, transcript):
|
24 |
+
model_ckpt = "deepset/minilm-uncased-squad2"
|
25 |
+
tokenizer = AutoTokenizer.from_pretrained(model_ckpt)
|
26 |
+
#question = "any funny examples in video??"
|
27 |
+
context = transcript
|
28 |
+
inputs = tokenizer(question, context, return_overflowing_tokens=True, max_length=512, stride = 25)
|
29 |
+
|
30 |
+
#overlaps
|
31 |
+
#getting a list of contexts available after striding
|
32 |
+
contx=[]
|
33 |
+
for window in inputs["input_ids"]:
|
34 |
+
#print(f"{tokenizer.decode(window)} \n")
|
35 |
+
contx.append(tokenizer.decode(window).split('[SEP]')[1].strip())
|
36 |
+
#print(ques)
|
37 |
+
#print(contx)
|
38 |
+
|
39 |
+
model = AutoModelForQuestionAnswering.from_pretrained(model_ckpt)
|
40 |
+
lst=[]
|
41 |
+
pipe = pipeline("question-answering", model=model, tokenizer=tokenizer)
|
42 |
+
for contexts in contx:
|
43 |
+
#print(pipe(question=question, context=contexts))
|
44 |
+
lst.append(pipe(question=question, context=contexts))
|
45 |
+
|
46 |
+
lst_scores = [dicts['score'] for dicts in lst]
|
47 |
+
#print(lst_scores)
|
48 |
+
#getting highest and second highest scores
|
49 |
+
idxmax = lst_scores.index(max(lst_scores))
|
50 |
+
lst_scores.remove(max(lst_scores))
|
51 |
+
idxmax2 = lst_scores.index(max(lst_scores))
|
52 |
+
#idxmax, idxmax2
|
53 |
+
|
54 |
+
idxcont = lst[idxmax2]['start']
|
55 |
+
answer = FinalTranscript[len(contx[0])-135 + idxcont:]
|
56 |
+
sentence_keyword = answer[:50]
|
57 |
+
|
58 |
+
dftranscript = pd.DataFrame(transcript)
|
59 |
+
#dftranscript.head()
|
60 |
+
|
61 |
+
modelST = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
|
62 |
+
embedding_1= modelST.encode(dftranscript.text, convert_to_tensor=True)
|
63 |
+
embedding_2 = modelST.encode(sentence_keyword, convert_to_tensor=True)
|
64 |
+
|
65 |
+
similarity_tensor = util.pytorch_cos_sim(embedding_1, embedding_2)
|
66 |
+
idx = torch.argmax(similarity_tensor)
|
67 |
+
start_timestamp = dftranscript.iloc[[int(idx)+1]].start.values[0]
|
68 |
+
start_timestamp = round(start_timestamp)
|
69 |
+
|
70 |
+
return start_timestamp
|
71 |
+
|
72 |
+
|
73 |
+
def display_vid(url, question):
|
74 |
#https://www.youtube.com/watch?v=smUHQndcmOY&t=425s
|
75 |
#html = HTML("<iframe width='560' height='315' src='https://www.youtube.com/watch?v=smUHQndcmOY&t=425s' frameborder='0' allowfullscreen></iframe>")
|
76 |
#html = "<iframe width='560' height='315' src='https://www.youtube.com/embed/smUHQndcmOY' frameborder='0' allowfullscreen></iframe>"
|
77 |
+
#get embedding and youtube link
|
78 |
+
html = "<iframe width='560' height='315' src=" + url + " frameborder='0' allowfullscreen></iframe>"
|
79 |
+
print(html)
|
80 |
+
|
81 |
+
#get transcript
|
82 |
+
transcript, video_id = get_transcript(html)
|
83 |
+
|
84 |
+
#get answer timestamp
|
85 |
+
#input - question and transcript, output - answer timestamp
|
86 |
+
ans_timestamp = get_answers_timestamp(question, transcript):
|
87 |
+
|
88 |
+
#created embedding
|
89 |
+
#sample - smUHQndcmOY?start=234
|
90 |
+
html_out = "<iframe width='560' height='315' src='https://www.youtube.com/embed/" + video_id + "?start=" + ans_timestamp + " title='YouTube video player' frameborder='0' allow='accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture' allowfullscreen></iframe>"
|
91 |
+
print(f"html output is : {html_out}")
|
92 |
#vid = YouTubeVideo('smUHQndcmOY&t=425s')
|
93 |
return html
|
94 |
+
|
95 |
#https://youtu.be/smUHQndcmOY
|
96 |
def fun(url):
|
97 |
return gr.Video(value=url)
|
|
|
108 |
)
|
109 |
with gr.Row():
|
110 |
input_url = gr.Textbox() #gr.HTML(placeholder="Enter a video link here..")
|
111 |
+
input_ques = gr.Textbox()
|
112 |
output_vid = gr.HTML()
|
113 |
|
114 |
b1 = gr.Button("Publish Video")
|
115 |
#b2 = gr.Button("Generate Image")
|
116 |
|
117 |
+
b1.click(display_vid, inputs=[input_url,input_ques], outputs=output_vid)
|
118 |
#b2.click(poem_to_image, poem_txt, output_image)
|
119 |
#examples=examples
|
120 |
|