TogetherAI commited on
Commit
1bfe127
1 Parent(s): 231f5c6

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +11 -25
app.py CHANGED
@@ -1,12 +1,11 @@
1
  import torch
2
-
3
  import gradio as gr
4
  import yt_dlp as youtube_dl
5
  from transformers import pipeline
6
  from transformers.pipelines.audio_utils import ffmpeg_read
7
-
8
  import tempfile
9
  import os
 
10
 
11
  MODEL_NAME = "openai/whisper-large-v3"
12
  BATCH_SIZE = 8
@@ -22,14 +21,12 @@ pipe = pipeline(
22
  device=device,
23
  )
24
 
25
-
26
  def transcribe(inputs, task):
27
  if inputs is None:
28
  raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")
29
 
30
  text = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)["text"]
31
- return text
32
-
33
 
34
  def _return_yt_html_embed(yt_url):
35
  video_id = yt_url.split("?v=")[-1]
@@ -47,19 +44,10 @@ def download_yt_audio(yt_url, filename):
47
  except youtube_dl.utils.DownloadError as err:
48
  raise gr.Error(str(err))
49
 
50
- file_length = info["duration_string"]
51
- file_h_m_s = file_length.split(":")
52
- file_h_m_s = [int(sub_length) for sub_length in file_h_m_s]
53
-
54
- if len(file_h_m_s) == 1:
55
- file_h_m_s.insert(0, 0)
56
- if len(file_h_m_s) == 2:
57
- file_h_m_s.insert(0, 0)
58
- file_length_s = file_h_m_s[0] * 3600 + file_h_m_s[1] * 60 + file_h_m_s[2]
59
-
60
- if file_length_s > YT_LENGTH_LIMIT_S:
61
  yt_length_limit_hms = time.strftime("%HH:%MM:%SS", time.gmtime(YT_LENGTH_LIMIT_S))
62
- file_length_hms = time.strftime("%HH:%MM:%SS", time.gmtime(file_length_s))
63
  raise gr.Error(f"Maximum YouTube length is {yt_length_limit_hms}, got {file_length_hms} YouTube video.")
64
 
65
  ydl_opts = {"outtmpl": filename, "format": "worstvideo[ext=mp4]+bestaudio[ext=m4a]/best[ext=mp4]/best"}
@@ -70,8 +58,7 @@ def download_yt_audio(yt_url, filename):
70
  except youtube_dl.utils.ExtractorError as err:
71
  raise gr.Error(str(err))
72
 
73
-
74
- def yt_transcribe(yt_url, task, max_filesize=75.0):
75
  html_embed_str = _return_yt_html_embed(yt_url)
76
 
77
  with tempfile.TemporaryDirectory() as tmpdirname:
@@ -87,8 +74,7 @@ def yt_transcribe(yt_url, task, max_filesize=75.0):
87
 
88
  return html_embed_str, text
89
 
90
-
91
- demo = gr.Blocks(theme="TogetherAI/Alex2_Theme")
92
 
93
  mf_transcribe = gr.Interface(
94
  fn=transcribe,
@@ -98,7 +84,7 @@ mf_transcribe = gr.Interface(
98
  ],
99
  outputs="text",
100
  layout="horizontal",
101
- theme="TogetherAI/Alex2_Theme",
102
  title="Whisper Large V3: Transcribe Audio",
103
  description=(
104
  "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the"
@@ -116,7 +102,7 @@ file_transcribe = gr.Interface(
116
  ],
117
  outputs="text",
118
  layout="horizontal",
119
- theme="TogetherAI/Alex2_Theme",
120
  title="Whisper Large V3: Transcribe Audio",
121
  description=(
122
  "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the"
@@ -134,7 +120,7 @@ yt_transcribe = gr.Interface(
134
  ],
135
  outputs=["html", "text"],
136
  layout="horizontal",
137
- theme="TogetherAI/Alex2_Theme",
138
  title="Whisper Large V3: Transcribe YouTube",
139
  description=(
140
  "Transcribe long-form YouTube videos with the click of a button! Demo uses the checkpoint"
@@ -147,4 +133,4 @@ yt_transcribe = gr.Interface(
147
  with demo:
148
  gr.TabbedInterface([mf_transcribe, file_transcribe, yt_transcribe], ["Microphone", "Audio file", "YouTube"])
149
 
150
- demo.launch(enable_queue=True)
 
1
  import torch
 
2
  import gradio as gr
3
  import yt_dlp as youtube_dl
4
  from transformers import pipeline
5
  from transformers.pipelines.audio_utils import ffmpeg_read
 
6
  import tempfile
7
  import os
8
+ import time
9
 
10
  MODEL_NAME = "openai/whisper-large-v3"
11
  BATCH_SIZE = 8
 
21
  device=device,
22
  )
23
 
 
24
  def transcribe(inputs, task):
25
  if inputs is None:
26
  raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")
27
 
28
  text = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)["text"]
29
+ return text
 
30
 
31
  def _return_yt_html_embed(yt_url):
32
  video_id = yt_url.split("?v=")[-1]
 
44
  except youtube_dl.utils.DownloadError as err:
45
  raise gr.Error(str(err))
46
 
47
+ file_length = info["duration"]
48
+ if file_length > YT_LENGTH_LIMIT_S:
 
 
 
 
 
 
 
 
 
49
  yt_length_limit_hms = time.strftime("%HH:%MM:%SS", time.gmtime(YT_LENGTH_LIMIT_S))
50
+ file_length_hms = time.strftime("%HH:%MM:%SS", time.gmtime(file_length))
51
  raise gr.Error(f"Maximum YouTube length is {yt_length_limit_hms}, got {file_length_hms} YouTube video.")
52
 
53
  ydl_opts = {"outtmpl": filename, "format": "worstvideo[ext=mp4]+bestaudio[ext=m4a]/best[ext=mp4]/best"}
 
58
  except youtube_dl.utils.ExtractorError as err:
59
  raise gr.Error(str(err))
60
 
61
+ def yt_transcribe(yt_url, task):
 
62
  html_embed_str = _return_yt_html_embed(yt_url)
63
 
64
  with tempfile.TemporaryDirectory() as tmpdirname:
 
74
 
75
  return html_embed_str, text
76
 
77
+ demo = gr.Blocks(theme="ParityError/Interstellar")
 
78
 
79
  mf_transcribe = gr.Interface(
80
  fn=transcribe,
 
84
  ],
85
  outputs="text",
86
  layout="horizontal",
87
+ theme="ParityError/Interstellar",
88
  title="Whisper Large V3: Transcribe Audio",
89
  description=(
90
  "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the"
 
102
  ],
103
  outputs="text",
104
  layout="horizontal",
105
+ theme="ParityError/Interstellar",
106
  title="Whisper Large V3: Transcribe Audio",
107
  description=(
108
  "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the"
 
120
  ],
121
  outputs=["html", "text"],
122
  layout="horizontal",
123
+ theme="ParityError/Interstellar",
124
  title="Whisper Large V3: Transcribe YouTube",
125
  description=(
126
  "Transcribe long-form YouTube videos with the click of a button! Demo uses the checkpoint"
 
133
  with demo:
134
  gr.TabbedInterface([mf_transcribe, file_transcribe, yt_transcribe], ["Microphone", "Audio file", "YouTube"])
135
 
136
+ demo.launch(enable_queue=True)