File size: 2,124 Bytes
c1cf3f5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5b0ba96
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
import gradio as gr
import whisper
from pytube import YouTube

def get_audio(url):
    yt = YouTube(url)
    return yt.streams.filter(only_audio=True)[0].download(filename="tmp.mp4")

def get_transcript(url, model_size, lang, format):

    model = whisper.load_model(model_size)

    if lang == "None":
        lang = None
    
    result = model.transcribe(get_audio(url), fp16=False, language=lang)

    if format == "None":
        return result["text"]
    elif format == ".srt":
        return format_to_srt(result["segments"])

def format_to_srt(segments):
    output = ""
    for i, segment in enumerate(segments):
        output += f"{i + 1}\n"
        output += f"{format_timestamp(segment['start'])} --> {format_timestamp(segment['end'])}\n"
        output += f"{segment['text']}\n\n"
    return output

def format_timestamp(t):
    hh = t//3600
    mm = (t - hh*3600)//60
    ss = t - hh*3600 - mm*60
    mi = (t - int(t))*1000
    return f"{int(hh):02d}:{int(mm):02d}:{int(ss):02d},{int(mi):03d}"


langs = ["None"] + sorted(list(whisper.tokenizer.LANGUAGES.values()))
model_size = list(whisper._MODELS.keys())

with gr.Blocks() as demo:

    with gr.Row():

        with gr.Column():

            with gr.Row():
                url = gr.Textbox(placeholder='Youtube video URL', label='URL')

            with gr.Row():

                model_size = gr.Dropdown(choices=model_size, value='tiny', label="Model")
                lang = gr.Dropdown(choices=langs, value="None", label="Language (Optional)")
                format = gr.Dropdown(choices=["None", ".srt"], value="None", label="Timestamps? (Optional)")

            with gr.Row():
                gr.Markdown("Larger models are more accurate, but slower. For 1min video, it'll take ~30s (tiny), ~1min (base), ~3min (small), ~5min (medium), etc.")
                transcribe_btn = gr.Button('Transcribe')

        with gr.Column():
            outputs = gr.Textbox(placeholder='Transcription of the video', label='Transcription')

    transcribe_btn.click(get_transcript, inputs=[url, model_size, lang, format], outputs=outputs)

demo.launch(debug=True)