Spaces:
Runtime error
Runtime error
Jiedong Yang
commited on
Commit
β’
22bdedf
1
Parent(s):
4591b06
Upload 2 files
Browse files- app.py +143 -0
- requirements.txt +16 -0
app.py
ADDED
@@ -0,0 +1,143 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import pafy
|
3 |
+
import time
|
4 |
+
import whisper
|
5 |
+
import validators
|
6 |
+
import gradio as gr
|
7 |
+
|
8 |
+
from wordcloud import WordCloud, STOPWORDS
|
9 |
+
|
10 |
+
# load whisper model for ASR and BART for summarization
|
11 |
+
asr_model = whisper.load_model('base.en')
|
12 |
+
summarizer = gr.Interface.load("facebook/bart-large-cnn", src='huggingface')
|
13 |
+
|
14 |
+
|
15 |
+
def audio_from_url(url, dst_dir='data', name=None, format='wav'):
|
16 |
+
""" Download video from url and save the audio from video
|
17 |
+
|
18 |
+
:param url: str, the video url
|
19 |
+
:param dst_dir: destination directory for save audio
|
20 |
+
:param name: audio file's name, if none, assign the name as the video's title
|
21 |
+
:param format: format type for audio file, such as 'wav', 'mp3'. WAV is preferred.
|
22 |
+
:return: path of audio
|
23 |
+
"""
|
24 |
+
|
25 |
+
if not validators.url(url):
|
26 |
+
return None
|
27 |
+
|
28 |
+
os.makedirs(dst_dir, exist_ok=True)
|
29 |
+
|
30 |
+
# download audio
|
31 |
+
video = pafy.new(url)
|
32 |
+
path = os.path.join(dst_dir, f"audio.{format}")
|
33 |
+
os.remove(path)
|
34 |
+
os.system(f"yt-dlp -f 'ba' -x --audio-format {format} {url} -o {path} --quiet")
|
35 |
+
|
36 |
+
return path
|
37 |
+
|
38 |
+
|
39 |
+
def speech_to_text(audio, beam_size=5, best_of=5, language='en'):
|
40 |
+
""" ASR inference with Whisper
|
41 |
+
|
42 |
+
:param audio:
|
43 |
+
:param beam_size:
|
44 |
+
:param best_of:
|
45 |
+
:param language:
|
46 |
+
:return:
|
47 |
+
"""
|
48 |
+
|
49 |
+
result = asr_model.transcribe(audio, language=language, beam_size=beam_size, best_of=best_of)
|
50 |
+
|
51 |
+
return result['text']
|
52 |
+
|
53 |
+
|
54 |
+
def text_summarization(text):
|
55 |
+
return summarizer(text)
|
56 |
+
|
57 |
+
|
58 |
+
def wordcloud_func(text: str, out_path='wordcloud_output.png'):
|
59 |
+
""" generate wordcloud based on text
|
60 |
+
|
61 |
+
:param text:
|
62 |
+
:param out_path:
|
63 |
+
:return:
|
64 |
+
"""
|
65 |
+
|
66 |
+
if len(text) == 0:
|
67 |
+
return None
|
68 |
+
|
69 |
+
stopwords = STOPWORDS
|
70 |
+
|
71 |
+
wc = WordCloud(
|
72 |
+
background_color='white',
|
73 |
+
stopwords=stopwords,
|
74 |
+
height=600,
|
75 |
+
width=600
|
76 |
+
)
|
77 |
+
|
78 |
+
wc.generate(text)
|
79 |
+
wc.to_file(out_path)
|
80 |
+
|
81 |
+
return out_path
|
82 |
+
|
83 |
+
|
84 |
+
demo = gr.Blocks(title="Speech Summarization")
|
85 |
+
|
86 |
+
demo.encrypt = False
|
87 |
+
|
88 |
+
with demo:
|
89 |
+
# demo description
|
90 |
+
gr.Markdown("""
|
91 |
+
## Speech Summarization with Whisper
|
92 |
+
This space is intended to summarize a speech, a short one or long one, to save us sometime.
|
93 |
+
1. Type in a youtube URL or upload an audio file
|
94 |
+
2. Generate transcription with Whisper (Currently English Only)
|
95 |
+
3. Summarize the transcribed speech
|
96 |
+
4. A little wordcloud for you as well
|
97 |
+
""")
|
98 |
+
|
99 |
+
# data preparation
|
100 |
+
with gr.Row():
|
101 |
+
with gr.Column():
|
102 |
+
url = gr.Textbox(label="URL", placeholder="video url")
|
103 |
+
|
104 |
+
url_btn = gr.Button("clear")
|
105 |
+
url_btn.click(lambda x: '', inputs=url, outputs=url)
|
106 |
+
|
107 |
+
speech = gr.Audio(label="Speech", type="filepath")
|
108 |
+
|
109 |
+
url.change(audio_from_url, inputs=url, outputs=speech)
|
110 |
+
|
111 |
+
# ASR
|
112 |
+
text = gr.Textbox(label="Transcription", placeholder="transcription")
|
113 |
+
|
114 |
+
with gr.Row():
|
115 |
+
beam_size_slider = gr.Slider(1, 10, value=5, step=1, label="param: beam_size")
|
116 |
+
best_of_slider = gr.Slider(1, 10, value=5, step=1, label="param: best_of")
|
117 |
+
|
118 |
+
with gr.Row():
|
119 |
+
asr_clr_btn = gr.Button("clear")
|
120 |
+
asr_clr_btn.click(lambda x: '', inputs=text, outputs=text)
|
121 |
+
asr_btn = gr.Button("Recognize Speech")
|
122 |
+
asr_btn.click(speech_to_text, inputs=[speech, beam_size_slider, best_of_slider], outputs=text)
|
123 |
+
|
124 |
+
# summarization
|
125 |
+
summary = gr.Textbox(label="Summarization")
|
126 |
+
|
127 |
+
with gr.Row():
|
128 |
+
sum_clr_btn = gr.Button("clear")
|
129 |
+
sum_clr_btn.click(lambda x: '', inputs=summary, outputs=summary)
|
130 |
+
sum_btn = gr.Button("Summarize")
|
131 |
+
sum_btn.click(text_summarization, inputs=text, outputs=summary)
|
132 |
+
|
133 |
+
# wordcloud
|
134 |
+
image = gr.Image(label="wordcloud", show_label=False).style(height=400, width=400)
|
135 |
+
|
136 |
+
text.change(wordcloud_func, inputs=text, outputs=image)
|
137 |
+
|
138 |
+
examples = gr.Examples(examples=["https://www.youtube.com/watch?v=DuX4K4eeTz8",
|
139 |
+
"https://www.youtube.com/watch?v=nepOSEGHHCQ"],
|
140 |
+
inputs=[url])
|
141 |
+
|
142 |
+
if __name__ == '__main__':
|
143 |
+
demo.launch(share=True)
|
requirements.txt
ADDED
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
fastapi==0.85.1
|
2 |
+
ffmpeg-python==0.2.0
|
3 |
+
gradio==3.6
|
4 |
+
huggingface-hub==0.9.1
|
5 |
+
matplotlib==3.6.1
|
6 |
+
pafy==0.5.5
|
7 |
+
pandas==1.5.0
|
8 |
+
tokenizers==0.12.1
|
9 |
+
torch==1.12.1
|
10 |
+
tqdm==4.64.1
|
11 |
+
transformers==4.22.1
|
12 |
+
validators==0.20.0
|
13 |
+
whisper @ git+https://github.com/openai/whisper.git@8cf36f3508c9acd341a45eb2364239a3d81458b9
|
14 |
+
wordcloud==1.8.2.2
|
15 |
+
youtube-dl==2021.12.17
|
16 |
+
yt-dlp==2022.10.4
|