TranscendAI/TextBot.py at main · TranscendAI-Transformers/TranscendAI · GitHub

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
from transformers import pipeline, set_seed
import os
from pytube import YouTube
import json
import torch
import math


class TextBot:

    def __init__(self):
        self.transcribe_model = pipeline(model="facebook/wav2vec2-large-960h-lv60-self", device=0, framework="pt")
        self.summarizer_model = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6", device=0,
                                         framework="pt")
        self.nlp = pipeline('question-answering', model="deepset/roberta-large-squad2",
                            tokenizer="deepset/roberta-large-squad2", device=0, framework="pt")
        self.text_generator = pipeline('text-generation', model='gpt2-large', device=0)
        self.url = None
        self.title = None
        self.res = {}
        self.yt = None
        self.audio_format = '.wav'
        self.output_format = '.json'
        self.temp_location = './temp/'
        torch.cuda.empty_cache()

    def process_url(self, url):
        self.url = url
        self.yt = YouTube(self.url)
        title = self.yt.title
        title = title.replace('.', '')
        title = title.replace('(', '')
        title = title.replace(')', '')
        title = title.replace('?', '')
        title = title.replace('|', '')
        title = title.replace(':', '')
        print(title)
        title = title.replace(' ', '_')
        self.title = title

    def check_cache(self):
        prev_file = self.temp_location + self.title + self.output_format
        if os.path.isfile(prev_file):
            with open(prev_file) as json_file:
                self.res = json.load(json_file)
                return self.res
        else:
            return None

    def download_video(self):
        title = self.title + self.audio_format
        if not os.path.isfile(self.temp_location + title):
            video = self.yt.streams.filter(only_audio=True).first()
            out_file = video.download(output_path=self.temp_location)
            base, ext = os.path.splitext(out_file)
            title = self.temp_location + title
            try:
                os.rename(out_file, title)
            except FileExistsError:
                os.remove(title)
                os.rename(out_file, title)
            return title
        print('downloaded video->', title)
        print('location->', self.temp_location + title)
        return './temp/' + title

    def transcribe_audio(self):
        new_file = self.download_video()
        text = self.transcribe_model(new_file, chunk_length_s=60)['text']
        self.res['text'] = text.lower()
        print('transcribe done')
        torch.cuda.empty_cache()

    def generate_summary(self):
        prev = 0
        text = self.res['text']
        self.res['summary'] = []
        if len(text) > 3000:
            batches = math.ceil(len(text) / 3000)
            for i in range(1, batches + 1):
                summary = self.summarizer_model(text[prev:i * 3000], max_length=150, min_length=30, do_sample=False)
                self.res['summary'].append(summary[0]['summary_text'].lower())
                prev += 3000
        else:
            summary = self.summarizer_model(text, max_length=150, min_length=30, do_sample=False)
            self.res['summary'].append(summary[0]['summary_text'].lower())
        print('summarization done')
        torch.cuda.empty_cache()

    def summary(self,text):
        prev = 0

        summary = []
        if len(text) > 3000:
            batches = math.ceil(len(text) / 3000)
            for i in range(1, batches + 1):
                t_sum = self.summarizer_model(text[prev:i * 3000], max_length=150, min_length=30, do_sample=False)
                summary.append(t_sum[0]['summary_text'].lower())
                prev += 3000
        else:
            t_sum = self.summarizer_model(text, max_length=150, min_length=30, do_sample=False)
            summary.append(t_sum[0]['summary_text'].lower())
        print('summarization done')
        torch.cuda.empty_cache()
        return summary

    def store_output(self):
        with open(self.temp_location + self.title + self.output_format, "w") as outfile:
            json.dump(self.res, outfile)

    def ask_qa_bot(self, question, context):
        qa_input = {
            'question': question,
            'context': context
        }
        ans = self.nlp(qa_input)
        torch.cuda.empty_cache()
        return ans

    def text_generation(self, text, multiple):
        num = 1
        if multiple:
            num = 3
        resp = self.text_generator(text, max_length=150, num_return_sequences=num, top_k=0,
                                   temperature=0.8, do_sample=True, )
        torch.cuda.empty_cache()
        return resp

    def run_pipeline(self, url, t_only):
        self.process_url(url)
        cache_hit = self.check_cache()
        if cache_hit is not None:
            return cache_hit
        self.transcribe_audio()
        if not t_only:
            self.generate_summary()
        self.store_output()
        torch.cuda.empty_cache()
        return self.res