File size: 23,260 Bytes
e48aa5d
cb67dcf
e48aa5d
 
 
4e5a5be
e48aa5d
 
 
5ed2636
cb67dcf
e48aa5d
 
 
 
 
d84d90d
e48aa5d
 
 
 
 
 
 
 
 
 
 
cb67dcf
e48aa5d
86a552a
 
 
 
 
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
53f76b1
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
86a552a
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86a552a
 
 
 
 
 
31eb124
e48aa5d
86a552a
31eb124
e48aa5d
86a552a
31eb124
 
 
 
 
 
e48aa5d
86a552a
31eb124
86a552a
31eb124
 
 
e48aa5d
86a552a
31eb124
 
 
 
 
e48aa5d
 
 
 
 
 
86a552a
e48aa5d
86a552a
 
 
e48aa5d
86a552a
e48aa5d
86a552a
 
 
e48aa5d
86a552a
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
 
 
206efbc
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
3714046
e48aa5d
 
31eb124
e48aa5d
 
 
 
 
 
 
31eb124
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86a552a
e44d2cc
e48aa5d
 
31eb124
e48aa5d
 
86a552a
e48aa5d
 
 
 
 
 
 
86a552a
e48aa5d
 
 
 
 
 
86a552a
e48aa5d
 
 
 
31eb124
e48aa5d
 
 
 
 
 
86a552a
e48aa5d
86a552a
 
 
e48aa5d
86a552a
e48aa5d
86a552a
 
 
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86a552a
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3f534ce
 
 
 
549e47a
3f534ce
 
 
 
 
 
 
 
 
 
 
549e47a
3f534ce
 
 
 
 
 
549e47a
3f534ce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
549e47a
3f534ce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
549e47a
3f534ce
 
 
549e47a
 
 
 
 
 
3f534ce
e48aa5d
 
59b2eeb
e48aa5d
d8e1fc4
e48aa5d
317f910
e48aa5d
 
 
 
 
 
8cc73eb
e48aa5d
034b845
e48aa5d
 
 
 
 
549e47a
e48aa5d
4faf3df
e48aa5d
 
 
 
 
 
 
 
 
 
 
cfdce4a
e48aa5d
cfdce4a
 
c1b89d5
cfdce4a
 
 
 
 
034b845
cfdce4a
e48aa5d
 
 
 
3be0656
 
 
e48aa5d
 
3be0656
 
 
e48aa5d
 
 
ae4aeee
8cc73eb
 
 
 
 
 
 
 
 
 
 
 
e48aa5d
 
4faf3df
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dadf984
e48aa5d
 
 
3be0656
e48aa5d
 
3be0656
e48aa5d
 
 
 
034b845
e48aa5d
4faf3df
14cf752
034b845
e48aa5d
 
 
 
 
 
3f534ce
4faf3df
3f534ce
 
 
 
 
 
dadf984
3f534ce
 
 
3be0656
3f534ce
 
3be0656
3f534ce
 
 
 
034b845
 
 
07d1dca
3f534ce
b8e3183
2be8bbf
3f534ce
 
e48aa5d
 
 
e44d2cc
e48aa5d
 
8c47083
 
 
5b87039
e48aa5d
 
 
 
 
 
 
 
 
 
 
cb67dcf
e48aa5d
 
 
 
e80f947
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9d6a48d
e48aa5d
 
 
 
 
 
 
 
 
 
 
 
cb67dcf
034b845
e48aa5d
761feb6
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
# Imports
import gradio as gr
import whisper
from pytube import YouTube
from transformers import pipeline, T5Tokenizer, T5ForConditionalGeneration, AutoTokenizer, AutoModelForSeq2SeqLM
import torch
from wordcloud import WordCloud
import re
import os


class GradioInference:
    def __init__(self):
        
        # OpenAI's Whisper model sizes
        self.sizes = list(whisper._MODELS.keys())

        # Whisper's available languages for ASR
        self.langs = ["none"] + sorted(list(whisper.tokenizer.LANGUAGES.values()))
        
        # Default size
        self.current_size = "base"
        
        # Default model size
        self.loaded_model = whisper.load_model(self.current_size)
        
        # Initialize Pytube Object
        self.yt = None

        # Initialize summary model for English
        self.bart_summarizer = pipeline("summarization", model="facebook/bart-large-cnn", truncation=True)

        # Initialize Multilingual summary model 
        self.mt5_tokenizer = AutoTokenizer.from_pretrained("csebuetnlp/mT5_multilingual_XLSum", truncation=True)
        self.mt5_model = AutoModelForSeq2SeqLM.from_pretrained("csebuetnlp/mT5_multilingual_XLSum")

        # Initialize VoiceLabT5 model and tokenizer
        self.keyword_model = T5ForConditionalGeneration.from_pretrained(
            "Voicelab/vlt5-base-keywords"
        )
        self.keyword_tokenizer = T5Tokenizer.from_pretrained(
            "Voicelab/vlt5-base-keywords"
        )

        # Sentiment Classifier
        self.classifier = pipeline("text-classification", model="lxyuan/distilbert-base-multilingual-cased-sentiments-student", return_all_scores=False)

    
    def __call__(self, link, lang, size, progress=gr.Progress()):
        """
        Call the Gradio Inference python class.
        This class gets access to a YouTube video using python's library Pytube and downloads its audio.
        Then it uses the Whisper model to perform Automatic Speech Recognition (i.e Speech-to-Text).
        Once the function has the transcription of the video it proccess it to obtain:
            - Summary: using Facebook's BART transformer.
            - KeyWords: using VoiceLabT5 keyword extractor.
            - Sentiment Analysis: using Hugging Face's default sentiment classifier
            - WordCloud: using the wordcloud python library.
        """
        progress(0, desc="Starting analysis")
        
        if self.yt is None:
            self.yt = YouTube(link)
        
        # Pytube library to access to YouTube audio stream
        path = self.yt.streams.filter(only_audio=True)[0].download(filename="tmp.mp4")

        if lang == "none":
            lang = None

        if size != self.current_size:
            self.loaded_model = whisper.load_model(size)
            self.current_size = size

        progress(0.20, desc="Transcribing")
        
        # Transcribe the audio extracted from pytube
        results = self.loaded_model.transcribe(path, language=lang)

        progress(0.40, desc="Summarizing")
        
        # Perform summarization on the transcription
        transcription_summary = self.bart_summarizer(
            results["text"], 
            max_length=256, 
            min_length=30, 
            do_sample=False, 
            truncation=True
        )

        #### Resumen multilingue con mt5 
        WHITESPACE_HANDLER = lambda k: re.sub('\s+', ' ', re.sub('\n+', ' ', k.strip()))
        
        input_ids_sum = self.mt5_tokenizer(
            [WHITESPACE_HANDLER(results["text"])],
            return_tensors="pt",
            padding="max_length",
            truncation=True,
            max_length=512
        )["input_ids"]
        
        output_ids_sum = self.mt5_model.generate(
            input_ids=input_ids_sum,
            max_length=256,
            no_repeat_ngram_size=2,
            num_beams=4
        )[0]
        
        summary = self.mt5_tokenizer.decode(
            output_ids_sum,
            skip_special_tokens=True,
            clean_up_tokenization_spaces=False
        )
        #### Fin resumen multilingue
        
        progress(0.60, desc="Extracting Keywords")
        
        # Extract keywords using VoiceLabT5
        task_prefix = "Keywords: "
        input_sequence = task_prefix + results["text"]
        
        input_ids = self.keyword_tokenizer(
            input_sequence, 
            return_tensors="pt", 
            truncation=False
        ).input_ids
        
        output = self.keyword_model.generate(
            input_ids, 
            no_repeat_ngram_size=3, 
            num_beams=4
        )
        
        predicted = self.keyword_tokenizer.decode(output[0], skip_special_tokens=True)
        keywords = [x.strip() for x in predicted.split(",") if x.strip()]
        formatted_keywords = "\n".join([f"β€’ {keyword}" for keyword in keywords])

        progress(0.80, desc="Extracting Sentiment")
       
        # Define a dictionary to map labels to emojis
        sentiment_emojis = {
            "positive": "Positive πŸ‘πŸΌ",
            "negative": "Negative πŸ‘ŽπŸΌ",
            "neutral": "Neutral 😢",
        }
        
        # Sentiment label    
        label = self.classifier(summary)[0]["label"]

        # Format the label with emojis
        formatted_sentiment = sentiment_emojis.get(label, label)

        progress(0.90, desc="Generating Wordcloud")
        
        # Generate WordCloud object
        wordcloud = WordCloud(colormap = "Oranges").generate(results["text"])

        # WordCloud image to display
        wordcloud_image = wordcloud.to_image()

        if lang == "english" or lang == "none":
            return (
                results["text"],
                transcription_summary[0]["summary_text"],
                formatted_keywords,
                formatted_sentiment,
                wordcloud_image,
            )
        else:
            return (
                results["text"],
                summary,
                formatted_keywords,
                formatted_sentiment,
                wordcloud_image,
            )


    def populate_metadata(self, link):
        """
        Access to the YouTube video title and thumbnail image to further display it
        params:
        - link: a YouTube URL.
        """
        if not link:
            return None, None
            
        self.yt = YouTube(link)
        return self.yt.thumbnail_url, self.yt.title

    def from_audio_input(self, lang, size, audio_file, progress=gr.Progress()):
        """
        Call the Gradio Inference python class.
        Uses it directly the Whisper model to perform Automatic Speech Recognition (i.e Speech-to-Text).
        Once the function has the transcription of the video it proccess it to obtain:
            - Summary: using Facebook's BART transformer.
            - KeyWords: using VoiceLabT5 keyword extractor.
            - Sentiment Analysis: using Hugging Face's default sentiment classifier
            - WordCloud: using the wordcloud python library.
        """
        progress(0, desc="Starting analysis")
        
        if lang == "none":
            lang = None

        if size != self.current_size:
            self.loaded_model = whisper.load_model(size)
            self.current_size = size

        progress(0.20, desc="Transcribing")
        
        results = self.loaded_model.transcribe(audio_file, language=lang)

        progress(0.40, desc="Summarizing")
        
        # Perform summarization on the transcription
        transcription_summary = self.bart_summarizer(
            results["text"], max_length=150, min_length=30, do_sample=False, truncation=True
        )
        
        #### Resumen multilingue
        WHITESPACE_HANDLER = lambda k: re.sub('\s+', ' ', re.sub('\n+', ' ', k.strip()))
        
        input_ids_sum = self.mt5_tokenizer(
            [WHITESPACE_HANDLER(results["text"])],
            return_tensors="pt",
            padding="max_length",
            truncation=True,
            max_length=512
        )["input_ids"]
        
        output_ids_sum = self.mt5_model.generate(
            input_ids=input_ids_sum,
            max_length=130,
            no_repeat_ngram_size=2,
            num_beams=4
        )[0]
        
        summary = self.mt5_tokenizer.decode(
            output_ids_sum,
            skip_special_tokens=True,
            clean_up_tokenization_spaces=False
        )
        #### Fin resumen multilingue

        progress(0.50, desc="Extracting Keywords")
        
        # Extract keywords using VoiceLabT5
        task_prefix = "Keywords: "
        input_sequence = task_prefix + results["text"]
        
        input_ids = self.keyword_tokenizer(
            input_sequence, 
            return_tensors="pt", 
            truncation=False
        ).input_ids
        
        output = self.keyword_model.generate(
            input_ids, 
            no_repeat_ngram_size=3, 
            num_beams=4
        )
        predicted = self.keyword_tokenizer.decode(output[0], skip_special_tokens=True)
        keywords = [x.strip() for x in predicted.split(",") if x.strip()]
        formatted_keywords = "\n".join([f"β€’ {keyword}" for keyword in keywords])

        progress(0.80, desc="Extracting Sentiment")

        # Define a dictionary to map labels to emojis
        sentiment_emojis = {
            "positive": "Positive πŸ‘πŸΌ",
            "negative": "Negative πŸ‘ŽπŸΌ",
            "neutral": "Neutral 😢",
        }
        
        # Sentiment label    
        label = self.classifier(summary)[0]["label"]

        # Format the label with emojis
        formatted_sentiment = sentiment_emojis.get(label, label)
        
        progress(0.90, desc="Generating Wordcloud")
        # WordCloud object
        wordcloud = WordCloud(colormap = "Oranges").generate(
            results["text"]
        )
        wordcloud_image = wordcloud.to_image()

        if lang == "english" or lang == "none":
            return (
                results["text"],
                transcription_summary[0]["summary_text"],
                formatted_keywords,
                formatted_sentiment,
                wordcloud_image,
            )
        else:
            return (
                results["text"],
                summary,
                formatted_keywords,
                formatted_sentiment,
                wordcloud_image,
            )

    
    def from_article(self, article, progress=gr.Progress()):
        """
        Call the Gradio Inference python class.
        Acepts the user's text imput, then it performs: 
            - Summary: using Facebook's BART transformer.
            - KeyWords: using VoiceLabT5 keyword extractor.
            - Sentiment Analysis: using Hugging Face's default sentiment classifier
            - WordCloud: using the wordcloud python library.
        """
        progress(0, desc="Starting analysis")

        progress(0.30, desc="Summarizing")
        
        # Perform summarization on the transcription
        transcription_summary = self.bart_summarizer(
            article, max_length=150, min_length=30, do_sample=False, truncation=True
        )
        
        #### Resumen multilingue
        WHITESPACE_HANDLER = lambda k: re.sub('\s+', ' ', re.sub('\n+', ' ', k.strip()))
        
        input_ids_sum = self.mt5_tokenizer(
            [WHITESPACE_HANDLER(article)],
            return_tensors="pt",
            padding="max_length",
            truncation=True,
            max_length=512
        )["input_ids"]
        
        output_ids_sum = self.mt5_model.generate(
            input_ids=input_ids_sum,
            max_length=130,
            no_repeat_ngram_size=2,
            num_beams=4
        )[0]
        
        summary = self.mt5_tokenizer.decode(
            output_ids_sum,
            skip_special_tokens=True,
            clean_up_tokenization_spaces=False
        )
        #### Fin resumen multilingue

        progress(0.60, desc="Extracting Keywords")
        
        # Extract keywords using VoiceLabT5
        task_prefix = "Keywords: "
        input_sequence = task_prefix + article
        
        input_ids = self.keyword_tokenizer(
            input_sequence, 
            return_tensors="pt", 
            truncation=False
        ).input_ids
        
        output = self.keyword_model.generate(
            input_ids, 
            no_repeat_ngram_size=3, 
            num_beams=4
        )
        predicted = self.keyword_tokenizer.decode(output[0], skip_special_tokens=True)
        keywords = [x.strip() for x in predicted.split(",") if x.strip()]
        formatted_keywords = "\n".join([f"β€’ {keyword}" for keyword in keywords])

        progress(0.80, desc="Extracting Sentiment")

        # Define a dictionary to map labels to emojis
        sentiment_emojis = {
            "positive": "Positive πŸ‘πŸΌ",
            "negative": "Negative πŸ‘ŽπŸΌ",
            "neutral": "Neutral 😢",
        }
        
        # Sentiment label    
        label = self.classifier(summary)[0]["label"]

        # Format the label with emojis
        formatted_sentiment = sentiment_emojis.get(label, label)
        
        progress(0.90, desc="Generating Wordcloud")
        # WordCloud object
        wordcloud = WordCloud(colormap = "Oranges").generate(
            article
        )
        wordcloud_image = wordcloud.to_image()

        return (
            transcription_summary[0]["summary_text"],
            formatted_keywords,
            formatted_sentiment,
            wordcloud_image,
        )


gio = GradioInference()
title = "Media Insights"
description = "Your AI-powered video analytics tool"
theme = gr.themes.Soft(spacing_size="lg", radius_size="lg")

block = gr.Blocks(theme=theme)

with block as demo:
    gr.HTML(
        """
        <div style="text-align: center; max-width: 500px; margin: 0 auto;">
          <div>
            <h1 style="font-family: Montserrat, sans-serif;">MEDIA <span style="color: #433ccb;">INSIGHTS</span> πŸ’‘</h1>
          </div>
          <h4 style="margin-bottom: 5px; font-size: 95%">
            Your AI-powered video analytics tool ✨
          </h4>
        </div>
        """
    )
    with gr.Group():
        with gr.Tab("From YouTube πŸ“Ή"):
            with gr.Box():
                
                with gr.Row().style(equal_height=True):
                    size = gr.Dropdown(
                        label="Speech-to-text Model Size", choices=gio.sizes, value="base"
                    )
                    lang = gr.Dropdown(
                        label="Language (Optional)", choices=gio.langs, value="none"
                    )
                link = gr.Textbox(
                    label="YouTube Link", placeholder="Enter YouTube link..."
                )

                with gr.Row().style(equal_height=True):
                    with gr.Column(variant="panel", scale=1):
                        title = gr.Label(label="Video Title")
                        img = gr.Image(label="Thumbnail").style(height=350)
                        
                    with gr.Column(variant="panel", scale=1):
                        text = gr.Textbox(
                            label="Transcription",
                            placeholder="Transcription Output...",
                            lines=18,
                        ).style(show_copy_button=True)
                    
                with gr.Row().style(equal_height=True):
                    summary = gr.Textbox(
                        label="Summary", placeholder="Summary Output...", lines=5
                    ).style(show_copy_button=True, 
                            # container=True
                           )
                    keywords = gr.Textbox(
                        label="Keywords", placeholder="Keywords Output...", lines=5
                    ).style(show_copy_button=True, 
                            # container=True
                           )
                    label = gr.Label(label="Sentiment Analysis")
                    wordcloud_image = gr.Image(label="WordCloud")
                    
                # with gr.Row():
            btn = gr.Button("Get Video Insights πŸ”Ž", label="Primary Button", variant="primary")
            clear = gr.ClearButton(
                [link, title, img, text, summary, keywords, label, wordcloud_image],
                label="Secondary Button",
                value="Clear πŸ—‘οΈ"
            )
            btn.click(
                gio,
                inputs=[link, lang, size],
                outputs=[text, summary, keywords, label, wordcloud_image],
            )
            link.change(gio.populate_metadata, inputs=[link], outputs=[img, title])

        with gr.Tab("From Audio file πŸŽ™οΈ"):
            with gr.Box():
                
                with gr.Row().style(equal_height=True):
                    size = gr.Dropdown(
                        label="Model Size", choices=gio.sizes, value="base"
                    )
                    lang = gr.Dropdown(
                        label="Language (Optional)", choices=gio.langs, value="none"
                    )
                audio_file = gr.Audio(type="filepath")
                
                with gr.Row().style(equal_height=True):
                    text = gr.Textbox(
                        label="Transcription",
                        placeholder="Transcription Output...",
                        lines=10,
                    ).style(show_copy_button=True)
                    
                with gr.Row().style(equal_height=True):
                    summary = gr.Textbox(
                        label="Summary", placeholder="Summary Output...", lines=5
                    )
                    keywords = gr.Textbox(
                        label="Keywords", placeholder="Keywords Output...", lines=5
                    )
                    label = gr.Label(label="Sentiment Analysis")
                    wordcloud_image = gr.Image(label="WordCloud")
                    
                with gr.Row():
                    btn = gr.Button(
                        "Get Audio Insights πŸ”Ž", variant="primary"
                    )
                    clear = gr.ClearButton([audio_file,text, summary, keywords, label, wordcloud_image], value="Clear πŸ—‘οΈ")
                btn.click(
                    gio.from_audio_input,
                    inputs=[lang, size, audio_file],
                    outputs=[text, summary, keywords, label, wordcloud_image],
                )

        with gr.Tab("From Article πŸ“‹"):
            with gr.Box():
                
                with gr.Row().style(equal_height=True):
                    article = gr.Textbox(
                        label="Transcription",
                        placeholder="Paste your text...",
                        lines=10,
                    ).style(show_copy_button=True)
                    
                with gr.Row().style(equal_height=True):
                    summary = gr.Textbox(
                        label="Summary", placeholder="Summary Output...", lines=5
                    )
                    keywords = gr.Textbox(
                        label="Keywords", placeholder="Keywords Output...", lines=5
                    )
                    label = gr.Label(label="Sentiment Analysis")
                    wordcloud_image = gr.Image(label="WordCloud")
                    
                with gr.Row(): 
                    btn = gr.Button(
                        "Get Text insights πŸ”Ž", variant="primary")
                    clear = gr.ClearButton([article, summary, keywords, label, wordcloud_image], value="Clear")
                btn.click(
                    gio.from_article,
                    inputs=[article],
                    outputs=[summary, keywords, label, wordcloud_image],
                )

with block:
    gr.Markdown("### Video Examples")
    gr.Examples(["https://www.youtube.com/shorts/xDNzz8yAH7I","https://www.youtube.com/watch?v=MnrJzXM7a6o&pp=ygURc3RldmUgam9icyBzcGVlY2g%3D"], inputs=link)

    gr.Markdown("### Audio Examples")
    # gr.Examples(
        # [[os.path.join(os.path.dirname(__file__),"audios/TED_lagrange_point.wav")],[os.path.join(os.path.dirname(__file__),"audios/TED_platon.wav")]], 
        # inputs=audio_file)
    
    gr.Markdown("### About the app:")

    with gr.Accordion("What is YouTube Insights?", open=False):
        gr.Markdown(
            "YouTube Insights is a tool developed for academic purposes that allows you to analyze YouTube videos or audio files. It provides features like transcription, summarization, keyword extraction, sentiment analysis, and word cloud generation for multimedia content."
        )

    with gr.Accordion("How does YouTube Insights work?", open=False):
        gr.Markdown(
            "YouTube Insights leverages several powerful AI models and libraries. It uses OpenAI's Whisper for Automatic Speech Recognition (ASR) to transcribe audio content. It summarizes the transcribed text using Facebook's BART model, extracts keywords with VoiceLabT5, performs sentiment analysis with DistilBERT, and generates word clouds."
        )

    with gr.Accordion("What languages are supported for the analysis?", open=False):
        gr.Markdown(
            "YouTube Insights supports multiple languages for transcription and analysis. You can select your preferred language from the available options when using the app."
        )

    with gr.Accordion("Can I analyze audio files instead of YouTube videos?", open=False):
        gr.Markdown(
            "Yes, you can analyze audio files directly. Simply upload your audio file to the app, and it will provide the same transcription, summarization, keyword extraction, sentiment analysis, and word cloud generation features."
        )

    with gr.Accordion("What are the different model sizes available for transcription?", open=False):
        gr.Markdown(
            "The app uses a Speech-to-text model that has different training sizes, from tiny to large. Hence, the bigger the model the accurate the transcription."
        )

    with gr.Accordion("How long does it take to analyze a video or audio file?", open=False):
        gr.Markdown(
            "The time taken for analysis may vary based on the duration of the video or audio file and the selected model size. Shorter content will be processed more quickly."
        )
    
    with gr.Accordion("Who developed YouTube Insights?" ,open=False):
        gr.Markdown(
            "YouTube Insights was developed by students as part of the 2022/23 Master's in Big Data & Data Science program at Universidad Complutense de Madrid for academic purposes (Trabajo de Fin de Master)."
        )
    
    gr.HTML(
        """
        <div style="text-align: center; max-width: 500px; margin: 0 auto;">
          <p style="margin-bottom: 10px; font-size: 96%">
            Trabajo de Fin de MΓ‘ster - Grupo 3
          </p>
          <p style="margin-bottom: 10px; font-size: 90%">
            2023 Master in Big Data & Data Science - Universidad Complutense de Madrid
          </p>
        </div>
        """
    )


demo.launch()