Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import AutoProcessor, SeamlessM4TModel
|
4 |
+
|
5 |
+
class SeamlessM4TApp:
|
6 |
+
def __init__(self):
|
7 |
+
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
8 |
+
print(f"Using device: {self.device}")
|
9 |
+
|
10 |
+
# Load model and processor
|
11 |
+
self.processor = AutoProcessor.from_pretrained("facebook/seamless-m4t-v2-large")
|
12 |
+
self.model = SeamlessM4TModel.from_pretrained("facebook/seamless-m4t-v2-large")
|
13 |
+
self.model.to(self.device)
|
14 |
+
|
15 |
+
def transcribe_audio(self, audio_path):
|
16 |
+
try:
|
17 |
+
# Load and process the audio
|
18 |
+
audio_inputs = self.processor(
|
19 |
+
audios=audio_path,
|
20 |
+
return_tensors="pt",
|
21 |
+
sampling_rate=16000
|
22 |
+
).to(self.device)
|
23 |
+
|
24 |
+
# Generate transcription
|
25 |
+
with torch.no_grad():
|
26 |
+
generated_tokens = self.model.generate(
|
27 |
+
**audio_inputs,
|
28 |
+
tgt_lang="eng",
|
29 |
+
task="transcribe"
|
30 |
+
)
|
31 |
+
|
32 |
+
# Decode the generated tokens
|
33 |
+
transcription = self.processor.decode(
|
34 |
+
generated_tokens[0].tolist(),
|
35 |
+
skip_special_tokens=True
|
36 |
+
)
|
37 |
+
|
38 |
+
return transcription
|
39 |
+
|
40 |
+
except Exception as e:
|
41 |
+
return f"Error during transcription: {str(e)}"
|
42 |
+
|
43 |
+
# Initialize the Gradio interface
|
44 |
+
def create_interface():
|
45 |
+
app = SeamlessM4TApp()
|
46 |
+
|
47 |
+
interface = gr.Interface(
|
48 |
+
fn=app.transcribe_audio,
|
49 |
+
inputs=gr.Audio(
|
50 |
+
type="filepath",
|
51 |
+
label="Upload Audio",
|
52 |
+
source="microphone"
|
53 |
+
),
|
54 |
+
outputs=gr.Textbox(label="Transcription"),
|
55 |
+
title="SeamlessM4T Speech-to-Text",
|
56 |
+
description="Upload audio or use microphone to transcribe speech to text using SeamlessM4T model.",
|
57 |
+
examples=[],
|
58 |
+
cache_examples=False
|
59 |
+
)
|
60 |
+
|
61 |
+
return interface
|
62 |
+
|
63 |
+
if __name__ == "__main__":
|
64 |
+
interface = create_interface()
|
65 |
+
interface.launch()
|