Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
|
3 |
+
import gradio as gr
|
4 |
+
from moviepy.editor import VideoFileClip
|
5 |
+
|
6 |
+
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
7 |
+
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
8 |
+
|
9 |
+
model_id = "distil-whisper/distil-large-v3"
|
10 |
+
|
11 |
+
model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
12 |
+
model_id, torch_dtype=torch_dtype, use_safetensors=True
|
13 |
+
)
|
14 |
+
model.to(device)
|
15 |
+
|
16 |
+
processor = AutoProcessor.from_pretrained(model_id)
|
17 |
+
|
18 |
+
pipe = pipeline(
|
19 |
+
"automatic-speech-recognition",
|
20 |
+
model=model,
|
21 |
+
tokenizer=processor.tokenizer,
|
22 |
+
feature_extractor=processor.feature_extractor,
|
23 |
+
max_new_tokens=128,
|
24 |
+
chunk_length_s=25,
|
25 |
+
batch_size=16,
|
26 |
+
torch_dtype=torch_dtype,
|
27 |
+
device=device,
|
28 |
+
)
|
29 |
+
|
30 |
+
def extract_audio_from_video(video_path, audio_output_path):
|
31 |
+
"""Extracts audio from a video and saves it to an MP3 file."""
|
32 |
+
try:
|
33 |
+
video_clip = VideoFileClip(video_path)
|
34 |
+
audio_clip = video_clip.audio
|
35 |
+
audio_clip.write_audiofile(audio_output_path)
|
36 |
+
print(f"Audio extracted successfully and saved to: {audio_output_path}")
|
37 |
+
return audio_output_path
|
38 |
+
except Exception as e:
|
39 |
+
print(f"Error extracting audio: {e}")
|
40 |
+
return None
|
41 |
+
|
42 |
+
def speech_to_text(input_file):
|
43 |
+
try:
|
44 |
+
if input_file.name.endswith((".mp4", ".avi", ".mov")):
|
45 |
+
audio_file_path = extract_audio_from_video(input_file.name, "temp_audio.mp3")
|
46 |
+
if audio_file_path:
|
47 |
+
result = pipe(audio_file_path)
|
48 |
+
return result[0]["transcription"]
|
49 |
+
else:
|
50 |
+
result = pipe(input_file.read())
|
51 |
+
return result[0]["transcription"]
|
52 |
+
except Exception as e:
|
53 |
+
return f"Error: {str(e)}"
|
54 |
+
|
55 |
+
iface = gr.Interface(fn=speech_to_text, inputs="file", outputs="text", title="Audio/Video-to-Text")
|
56 |
+
|
57 |
+
if __name__ == "__main__":
|
58 |
+
iface.launch(debug=True)
|