Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,148 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import whisper
|
3 |
+
import yt_dlp
|
4 |
+
import os
|
5 |
+
import traceback
|
6 |
+
from pydub import AudioSegment
|
7 |
+
from threading import Thread
|
8 |
+
from queue import Queue
|
9 |
+
|
10 |
+
# Global variable to store the selected model
|
11 |
+
selected_model = None
|
12 |
+
|
13 |
+
def load_whisper_model(model_name):
|
14 |
+
global selected_model
|
15 |
+
selected_model = whisper.load_model(model_name)
|
16 |
+
return f"Loaded {model_name} model"
|
17 |
+
|
18 |
+
def chunk_audio(audio_file, chunk_size_ms=30000):
|
19 |
+
audio = AudioSegment.from_file(audio_file)
|
20 |
+
chunks = [audio[i:i+chunk_size_ms] for i in range(0, len(audio), chunk_size_ms)]
|
21 |
+
return chunks
|
22 |
+
|
23 |
+
def stream_transcription(audio_file):
|
24 |
+
segment_queue = Queue()
|
25 |
+
|
26 |
+
def transcribe_worker():
|
27 |
+
try:
|
28 |
+
chunks = chunk_audio(audio_file)
|
29 |
+
for i, chunk in enumerate(chunks):
|
30 |
+
chunk_file = f"temp_chunk_{i}.wav"
|
31 |
+
chunk.export(chunk_file, format="wav")
|
32 |
+
result = selected_model.transcribe(chunk_file)
|
33 |
+
os.remove(chunk_file)
|
34 |
+
for segment in result['segments']:
|
35 |
+
segment_text = f"[{segment['start'] + i*30:.2f}s -> {segment['end'] + i*30:.2f}s] {segment['text']}\n"
|
36 |
+
segment_queue.put(segment_text)
|
37 |
+
segment_queue.put(None) # Signal end of transcription
|
38 |
+
except Exception as e:
|
39 |
+
segment_queue.put(f"Error: {str(e)}")
|
40 |
+
segment_queue.put(None)
|
41 |
+
|
42 |
+
Thread(target=transcribe_worker).start()
|
43 |
+
|
44 |
+
full_transcript = ""
|
45 |
+
while True:
|
46 |
+
segment_text = segment_queue.get()
|
47 |
+
if segment_text is None:
|
48 |
+
break
|
49 |
+
if segment_text.startswith("Error"):
|
50 |
+
yield segment_text
|
51 |
+
break
|
52 |
+
full_transcript += segment_text
|
53 |
+
yield full_transcript
|
54 |
+
|
55 |
+
def download_youtube_audio(youtube_url):
|
56 |
+
ydl_opts = {
|
57 |
+
'format': 'bestaudio/best',
|
58 |
+
'postprocessors': [{
|
59 |
+
'key': 'FFmpegExtractAudio',
|
60 |
+
'preferredcodec': 'mp3',
|
61 |
+
'preferredquality': '192',
|
62 |
+
}],
|
63 |
+
'outtmpl': 'temp_audio.%(ext)s',
|
64 |
+
}
|
65 |
+
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
66 |
+
ydl.download([youtube_url])
|
67 |
+
return "temp_audio.mp3"
|
68 |
+
|
69 |
+
def process_input(model, input_type, youtube_url=None, audio_file=None):
|
70 |
+
try:
|
71 |
+
yield "Loading Whisper model..."
|
72 |
+
load_whisper_model(model)
|
73 |
+
yield f"Loaded {model} model. "
|
74 |
+
|
75 |
+
if input_type == "YouTube URL":
|
76 |
+
if youtube_url:
|
77 |
+
yield "Downloading audio from YouTube..."
|
78 |
+
audio_file = download_youtube_audio(youtube_url)
|
79 |
+
yield "Download complete. Starting transcription...\n"
|
80 |
+
else:
|
81 |
+
yield "Please provide a valid YouTube URL."
|
82 |
+
return
|
83 |
+
elif input_type == "Audio File":
|
84 |
+
if not audio_file:
|
85 |
+
yield "Please upload an audio file."
|
86 |
+
return
|
87 |
+
else:
|
88 |
+
yield "Starting transcription...\n"
|
89 |
+
|
90 |
+
yield from stream_transcription(audio_file)
|
91 |
+
except Exception as e:
|
92 |
+
error_msg = f"An error occurred: {str(e)}\n"
|
93 |
+
error_msg += traceback.format_exc()
|
94 |
+
print(error_msg)
|
95 |
+
yield f"Error: {str(e)}"
|
96 |
+
finally:
|
97 |
+
if input_type == "YouTube URL" and audio_file:
|
98 |
+
os.remove(audio_file)
|
99 |
+
# Define the Gradio interface
|
100 |
+
with gr.Blocks() as iface:
|
101 |
+
gr.Markdown("# Whisper Transcription App")
|
102 |
+
gr.Markdown("Transcribe YouTube videos or audio files using OpenAI's Whisper model. Large files and long videos can take a very long time to process.")
|
103 |
+
|
104 |
+
with gr.Row():
|
105 |
+
with gr.Column():
|
106 |
+
model = gr.Radio(
|
107 |
+
choices=["tiny", "base", "small", "medium", "large"],
|
108 |
+
label="Whisper Model",
|
109 |
+
value="base"
|
110 |
+
)
|
111 |
+
gr.Markdown("""
|
112 |
+
- tiny: very fast, less accurate
|
113 |
+
- base: medium speed and accuracy
|
114 |
+
- small: balanced speed and accuracy
|
115 |
+
- medium: more accurate, slower
|
116 |
+
- large: most accurate, very slow
|
117 |
+
""")
|
118 |
+
|
119 |
+
input_type = gr.Radio(["YouTube URL", "Audio File"], label="Input Type")
|
120 |
+
youtube_url = gr.Textbox(label="YouTube URL")
|
121 |
+
audio_file = gr.Audio(label="Audio File", type="filepath")
|
122 |
+
|
123 |
+
with gr.Row():
|
124 |
+
submit_button = gr.Button("Submit")
|
125 |
+
clear_button = gr.Button("Clear")
|
126 |
+
|
127 |
+
with gr.Column():
|
128 |
+
output = gr.Textbox(label="Transcription", lines=25)
|
129 |
+
|
130 |
+
submit_button.click(
|
131 |
+
fn=process_input,
|
132 |
+
inputs=[model, input_type, youtube_url, audio_file],
|
133 |
+
outputs=output,
|
134 |
+
api_name="transcribe"
|
135 |
+
)
|
136 |
+
|
137 |
+
def clear_outputs():
|
138 |
+
return {youtube_url: "", audio_file: None, output: ""}
|
139 |
+
|
140 |
+
clear_button.click(
|
141 |
+
fn=clear_outputs,
|
142 |
+
inputs=[],
|
143 |
+
outputs=[youtube_url, audio_file, output],
|
144 |
+
api_name="clear"
|
145 |
+
)
|
146 |
+
|
147 |
+
# Launch the interface
|
148 |
+
iface.queue().launch(share=True)
|