devilent2 commited on
Commit
f6b3810
1 Parent(s): 3c1714f

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +10 -97
app.py CHANGED
@@ -1,99 +1,13 @@
1
- import spaces
2
- import torch
3
-
4
  import gradio as gr
5
- import yt_dlp as youtube_dl
6
- from transformers import pipeline
7
- from transformers.pipelines.audio_utils import ffmpeg_read
8
-
9
- import tempfile
10
- import os
11
-
12
- MODEL_NAME = "openai/whisper-large-v3"
13
- BATCH_SIZE = 8
14
- FILE_LIMIT_MB = 1000
15
- YT_LENGTH_LIMIT_S = 3600 # limit to 1 hour YouTube files
16
-
17
- device = 0 if torch.cuda.is_available() else "cpu"
18
-
19
- pipe = pipeline(
20
- task="automatic-speech-recognition",
21
- model=MODEL_NAME,
22
- chunk_length_s=30,
23
- device=device,
24
- )
25
-
26
-
27
- @spaces.GPU
28
- def transcribe(inputs, task):
29
- if inputs is None:
30
- raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")
31
-
32
- text = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)["text"]
33
- return text
34
-
35
-
36
- def _return_yt_html_embed(yt_url):
37
- video_id = yt_url.split("?v=")[-1]
38
- HTML_str = (
39
- f'<center> <iframe width="500" height="320" src="https://www.youtube.com/embed/{video_id}"> </iframe>'
40
- " </center>"
41
- )
42
- return HTML_str
43
-
44
- def download_yt_audio(yt_url, filename):
45
- info_loader = youtube_dl.YoutubeDL()
46
-
47
- try:
48
- info = info_loader.extract_info(yt_url, download=False)
49
- except youtube_dl.utils.DownloadError as err:
50
- raise gr.Error(str(err))
51
-
52
- file_length = info["duration_string"]
53
- file_h_m_s = file_length.split(":")
54
- file_h_m_s = [int(sub_length) for sub_length in file_h_m_s]
55
-
56
- if len(file_h_m_s) == 1:
57
- file_h_m_s.insert(0, 0)
58
- if len(file_h_m_s) == 2:
59
- file_h_m_s.insert(0, 0)
60
- file_length_s = file_h_m_s[0] * 3600 + file_h_m_s[1] * 60 + file_h_m_s[2]
61
-
62
- if file_length_s > YT_LENGTH_LIMIT_S:
63
- yt_length_limit_hms = time.strftime("%HH:%MM:%SS", time.gmtime(YT_LENGTH_LIMIT_S))
64
- file_length_hms = time.strftime("%HH:%MM:%SS", time.gmtime(file_length_s))
65
- raise gr.Error(f"Maximum YouTube length is {yt_length_limit_hms}, got {file_length_hms} YouTube video.")
66
-
67
- ydl_opts = {"outtmpl": filename, "format": "worstvideo[ext=mp4]+bestaudio[ext=m4a]/best[ext=mp4]/best"}
68
-
69
- with youtube_dl.YoutubeDL(ydl_opts) as ydl:
70
- try:
71
- ydl.download([yt_url])
72
- except youtube_dl.utils.ExtractorError as err:
73
- raise gr.Error(str(err))
74
-
75
- @spaces.GPU
76
- def yt_transcribe(yt_url, task, max_filesize=75.0):
77
- html_embed_str = _return_yt_html_embed(yt_url)
78
-
79
- with tempfile.TemporaryDirectory() as tmpdirname:
80
- filepath = os.path.join(tmpdirname, "video.mp4")
81
- download_yt_audio(yt_url, filepath)
82
- with open(filepath, "rb") as f:
83
- inputs = f.read()
84
-
85
- inputs = ffmpeg_read(inputs, pipe.feature_extractor.sampling_rate)
86
- inputs = {"array": inputs, "sampling_rate": pipe.feature_extractor.sampling_rate}
87
-
88
- text = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)["text"]
89
-
90
- return html_embed_str, text
91
-
92
 
93
  demo = gr.Blocks()
94
 
95
  mf_transcribe = gr.Interface(
96
- fn=transcribe,
97
  inputs=[
98
  gr.Audio(sources="microphone", type="filepath"),
99
  gr.Radio(["transcribe", "translate"], label="Task", value="transcribe"),
@@ -109,7 +23,7 @@ mf_transcribe = gr.Interface(
109
  )
110
 
111
  file_transcribe = gr.Interface(
112
- fn=transcribe,
113
  inputs=[
114
  gr.Audio(sources="upload", type="filepath", label="Audio file"),
115
  gr.Radio(["transcribe", "translate"], label="Task", value="transcribe"),
@@ -124,8 +38,8 @@ file_transcribe = gr.Interface(
124
  allow_flagging="never",
125
  )
126
 
127
- yt_transcribe = gr.Interface(
128
- fn=yt_transcribe,
129
  inputs=[
130
  gr.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL"),
131
  gr.Radio(["transcribe", "translate"], label="Task", value="transcribe")
@@ -141,7 +55,6 @@ yt_transcribe = gr.Interface(
141
  )
142
 
143
  with demo:
144
- gr.TabbedInterface([mf_transcribe, file_transcribe, yt_transcribe], ["Microphone", "Audio file", "YouTube"])
145
-
146
- demo.queue().launch()
147
 
 
 
 
 
 
1
  import gradio as gr
2
+ import spaces
3
+ from utils import MODEL_NAME
4
+ from transcribe import transcribe
5
+ from youtube import yt_transcribe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
 
7
  demo = gr.Blocks()
8
 
9
  mf_transcribe = gr.Interface(
10
+ fn=spaces.GPU(transcribe),
11
  inputs=[
12
  gr.Audio(sources="microphone", type="filepath"),
13
  gr.Radio(["transcribe", "translate"], label="Task", value="transcribe"),
 
23
  )
24
 
25
  file_transcribe = gr.Interface(
26
+ fn=spaces.GPU(transcribe),
27
  inputs=[
28
  gr.Audio(sources="upload", type="filepath", label="Audio file"),
29
  gr.Radio(["transcribe", "translate"], label="Task", value="transcribe"),
 
38
  allow_flagging="never",
39
  )
40
 
41
+ yt_transcribe_interface = gr.Interface(
42
+ fn=spaces.GPU(yt_transcribe),
43
  inputs=[
44
  gr.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL"),
45
  gr.Radio(["transcribe", "translate"], label="Task", value="transcribe")
 
55
  )
56
 
57
  with demo:
58
+ gr.TabbedInterface([mf_transcribe, file_transcribe, yt_transcribe_interface], ["Microphone", "Audio file", "YouTube"])
 
 
59
 
60
+ demo.queue().launch()