Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,11 +1,14 @@
|
|
1 |
import torch
|
2 |
|
3 |
import gradio as gr
|
4 |
-
import
|
|
|
5 |
from transformers import pipeline
|
6 |
from huggingface_hub import model_info
|
7 |
|
8 |
MODEL_NAME = "razhan/whisper-small-ckb"
|
|
|
|
|
9 |
|
10 |
device = 0 if torch.cuda.is_available() else "cpu"
|
11 |
pipe = pipeline(
|
@@ -36,23 +39,79 @@ def transcribe(microphone, file_upload):
|
|
36 |
|
37 |
|
38 |
def _return_yt_html_embed(yt_url):
|
39 |
-
|
|
|
|
|
|
|
|
|
40 |
HTML_str = (
|
41 |
-
f'<center
|
42 |
-
"
|
|
|
43 |
)
|
44 |
return HTML_str
|
45 |
|
46 |
|
47 |
-
def yt_transcribe(yt_url):
|
48 |
-
yt = pt.YouTube(yt_url)
|
49 |
-
html_embed_str = _return_yt_html_embed(yt_url)
|
50 |
-
stream = yt.streams.filter(only_audio=True)[0]
|
51 |
-
stream.download(filename="audio.mp3")
|
52 |
|
53 |
-
text = pipe("audio.mp3")["text"]
|
54 |
|
55 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
|
57 |
|
58 |
demo = gr.Blocks()
|
@@ -77,7 +136,14 @@ mf_transcribe = gr.Interface(
|
|
77 |
yt_transcribe = gr.Interface(
|
78 |
fn=yt_transcribe,
|
79 |
inputs=[gr.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL")],
|
80 |
-
outputs=["html",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
81 |
theme="huggingface",
|
82 |
title="Whisper Central Kurdish (Sorani) Demo: Transcribe YouTube",
|
83 |
description=(
|
|
|
1 |
import torch
|
2 |
|
3 |
import gradio as gr
|
4 |
+
import yt_dlp as youtube_dl
|
5 |
+
|
6 |
from transformers import pipeline
|
7 |
from huggingface_hub import model_info
|
8 |
|
9 |
MODEL_NAME = "razhan/whisper-small-ckb"
|
10 |
+
BATCH_SIZE = 1
|
11 |
+
FILE_LIMIT_MB = 10
|
12 |
|
13 |
device = 0 if torch.cuda.is_available() else "cpu"
|
14 |
pipe = pipeline(
|
|
|
39 |
|
40 |
|
41 |
def _return_yt_html_embed(yt_url):
|
42 |
+
if 'youtu.be' in yt_url:
|
43 |
+
video_id = yt_url.split('/')[-1].split('?')[0]
|
44 |
+
else:
|
45 |
+
video_id = yt_url.split("?v=")[-1].split('&')[0]
|
46 |
+
|
47 |
HTML_str = (
|
48 |
+
f'<center><iframe width="560" height="315" src="https://www.youtube.com/embed/{video_id}" '
|
49 |
+
'frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" '
|
50 |
+
'allowfullscreen></iframe></center>'
|
51 |
)
|
52 |
return HTML_str
|
53 |
|
54 |
|
|
|
|
|
|
|
|
|
|
|
55 |
|
|
|
56 |
|
57 |
+
|
58 |
+
def yt_transcribe(yt_url, task="transcribe", max_filesize=75.0, progress=gr.Progress()):
|
59 |
+
html_embed_str = _return_yt_html_embed(yt_url)
|
60 |
+
|
61 |
+
with tempfile.TemporaryDirectory() as tmpdirname:
|
62 |
+
filepath = os.path.join(tmpdirname, "video.mp4")
|
63 |
+
download_yt_audio(yt_url, filepath)
|
64 |
+
with open(filepath, "rb") as f:
|
65 |
+
inputs = f.read()
|
66 |
+
|
67 |
+
inputs = ffmpeg_read(inputs, pipe.feature_extractor.sampling_rate)
|
68 |
+
inputs = {"array": inputs, "sampling_rate": pipe.feature_extractor.sampling_rate}
|
69 |
+
|
70 |
+
|
71 |
+
|
72 |
+
start_time = time.time()
|
73 |
+
outputs = pipe(inputs, chunk_length_s=30, batch_size=BATCH_SIZE, generate_kwargs={"task": task, "language": "persian"}, return_timestamps=False)
|
74 |
+
exec_time = time.time() - start_time
|
75 |
+
logging.info(print(f"transcribe: {exec_time} sec."))
|
76 |
+
|
77 |
+
return html_embed_str, txt, exec_time
|
78 |
+
|
79 |
+
|
80 |
+
def download_yt_audio(yt_url, filename, progress=gr.Progress()):
|
81 |
+
if '&list' in yt_url:
|
82 |
+
yt_url = yt_url.split('&list')[0]
|
83 |
+
|
84 |
+
info_loader = youtube_dl.YoutubeDL()
|
85 |
+
|
86 |
+
try:
|
87 |
+
info = info_loader.extract_info(yt_url, download=False)
|
88 |
+
except youtube_dl.utils.DownloadError as err:
|
89 |
+
raise gr.Error(str(err))
|
90 |
+
|
91 |
+
file_length = info["duration_string"]
|
92 |
+
file_h_m_s = file_length.split(":")
|
93 |
+
file_h_m_s = [int(sub_length) for sub_length in file_h_m_s]
|
94 |
+
|
95 |
+
if len(file_h_m_s) == 1:
|
96 |
+
file_h_m_s.insert(0, 0)
|
97 |
+
if len(file_h_m_s) == 2:
|
98 |
+
file_h_m_s.insert(0, 0)
|
99 |
+
file_length_s = file_h_m_s[0] * 3600 + file_h_m_s[1] * 60 + file_h_m_s[2]
|
100 |
+
|
101 |
+
if file_length_s > YT_LENGTH_LIMIT_S:
|
102 |
+
yt_length_limit_hms = time.strftime("%HH:%MM:%SS", time.gmtime(YT_LENGTH_LIMIT_S))
|
103 |
+
file_length_hms = time.strftime("%HH:%MM:%SS", time.gmtime(file_length_s))
|
104 |
+
raise gr.Error(f"Maximum YouTube length is {yt_length_limit_hms}, got {file_length_hms} YouTube video.")
|
105 |
+
|
106 |
+
# ydl_opts = {"outtmpl": filename, "format": "worstvideo[ext=mp4]+bestaudio[ext=m4a]/best[ext=mp4]/best"}
|
107 |
+
ydl_opts = {"outtmpl": filename, "format": "bestaudio/best"}
|
108 |
+
|
109 |
+
with youtube_dl.YoutubeDL(ydl_opts) as ydl:
|
110 |
+
try:
|
111 |
+
ydl.download([yt_url])
|
112 |
+
except youtube_dl.utils.ExtractorError as err:
|
113 |
+
raise gr.Error(str(err))
|
114 |
+
progress(1, desc="Video downloaded from YouTube!")
|
115 |
|
116 |
|
117 |
demo = gr.Blocks()
|
|
|
136 |
yt_transcribe = gr.Interface(
|
137 |
fn=yt_transcribe,
|
138 |
inputs=[gr.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL")],
|
139 |
+
outputs=["html",
|
140 |
+
gr.Textbox(
|
141 |
+
label="Output",
|
142 |
+
rtl=True,
|
143 |
+
show_copy_button=True,
|
144 |
+
),
|
145 |
+
gr.Text(label="Transcription Time")
|
146 |
+
],
|
147 |
theme="huggingface",
|
148 |
title="Whisper Central Kurdish (Sorani) Demo: Transcribe YouTube",
|
149 |
description=(
|