Hugging Face's logo

Spaces:
jiedong-yang
/
Speech-Summarization-with-Whisper
Runtime error

App Files Community
Speech-Summarization-with-Whisper
File size: 4,062 Bytes 
22bdedf
09182d9
22bdedf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
09182d9
22bdedf
18d2fb0
 
22bdedf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
263d74d
22bdedf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
09182d9
 
 
 
22bdedf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
09182d9
22bdedf
 
e978023
 
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
 
import os
import pafy
import time
import whisper
import validators
import gradio as gr

from wordcloud import WordCloud, STOPWORDS

# load whisper model for ASR and BART for summarization
asr_model = whisper.load_model('base.en')
summarizer = gr.Interface.load("facebook/bart-large-cnn", src='huggingface')


def audio_from_url(url, dst_dir='data', name=None, format='wav'):
    """ Download video from url and save the audio from video

    :param url: str, the video url
    :param dst_dir: destination directory for save audio
    :param name: audio file's name, if none, assign the name as the video's title
    :param format: format type for audio file, such as 'wav', 'mp3'. WAV is preferred.
    :return: path of audio
    """

    if not validators.url(url):
        return None

    os.makedirs(dst_dir, exist_ok=True)

    # download audio
    # video = pafy.new(url)
    path = os.path.join(dst_dir, f"audio.{format}")
    if os.path.exists(path):
        os.remove(path)
    os.system(f"yt-dlp -f 'ba' -x --audio-format {format} {url}  -o {path} --quiet")

    return path


def speech_to_text(audio, beam_size=5, best_of=5, language='en'):
    """ ASR inference with Whisper

    :param audio:
    :param beam_size:
    :param best_of:
    :param language:
    :return:
    """

    result = asr_model.transcribe(audio, language=language, beam_size=beam_size, best_of=best_of, fp16=False)

    return result['text']


def text_summarization(text):
    return summarizer(text)


def wordcloud_func(text: str, out_path='wordcloud_output.png'):
    """ generate wordcloud based on text

    :param text:
    :param out_path:
    :return:
    """

    if len(text) == 0:
        return None

    stopwords = STOPWORDS

    wc = WordCloud(
        background_color='white',
        stopwords=stopwords,
        height=600,
        width=600
    )

    wc.generate(text)
    wc.to_file(out_path)

    return out_path


demo = gr.Blocks(title="Speech Summarization")

demo.encrypt = False

with demo:
    # demo description
    gr.Markdown("""
    ## Speech Summarization with Whisper
    This space is intended to summarize a speech, a short one or long one, to save us sometime.
    1. Type in a youtube URL or upload an audio file
    2. Generate transcription with Whisper (Currently English Only)
    3. Summarize the transcribed speech
    4. A little wordcloud for you as well
    """)

    # data preparation
    with gr.Row():
        with gr.Column():
            url = gr.Textbox(label="URL", placeholder="video url")

            url_btn = gr.Button("clear")
            url_btn.click(lambda x: '', inputs=url, outputs=url)

        speech = gr.Audio(label="Speech", type="filepath")

        url.change(audio_from_url, inputs=url, outputs=speech)

    examples = gr.Examples(examples=["https://www.youtube.com/watch?v=DuX4K4eeTz8",
                                     "https://www.youtube.com/watch?v=nepOSEGHHCQ"],
                           inputs=[url])

    # ASR
    text = gr.Textbox(label="Transcription", placeholder="transcription")

    with gr.Row():
        beam_size_slider = gr.Slider(1, 10, value=5, step=1, label="param: beam_size")
        best_of_slider = gr.Slider(1, 10, value=5, step=1, label="param: best_of")

    with gr.Row():
        asr_clr_btn = gr.Button("clear")
        asr_clr_btn.click(lambda x: '', inputs=text, outputs=text)
        asr_btn = gr.Button("Recognize Speech")
        asr_btn.click(speech_to_text, inputs=[speech, beam_size_slider, best_of_slider], outputs=text)

    # summarization
    summary = gr.Textbox(label="Summarization")

    with gr.Row():
        sum_clr_btn = gr.Button("clear")
        sum_clr_btn.click(lambda x: '', inputs=summary, outputs=summary)
        sum_btn = gr.Button("Summarize")
        sum_btn.click(text_summarization, inputs=text, outputs=summary)

    # wordcloud
    image = gr.Image(label="wordcloud", show_label=False).style(height=400, width=400)

    text.change(wordcloud_func, inputs=text, outputs=image)
    

if __name__ == '__main__':
    demo.launch()