Commit
·
cfa5958
1
Parent(s):
978917b
Add requirements.txt, Modifiy app.py
Browse files- app.py +110 -4
- requirements.txt +5 -0
app.py
CHANGED
@@ -1,7 +1,113 @@
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
|
|
|
|
2 |
|
3 |
-
def greet(name):
|
4 |
-
return "Hello " + name + "!!"
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import whisper
|
3 |
+
import torchaudio
|
4 |
import gradio as gr
|
5 |
+
import torch.nn as nn
|
6 |
+
from huggingface_hub import hf_hub_download
|
7 |
|
|
|
|
|
8 |
|
9 |
+
# Define the same model class used during training
|
10 |
+
class DialectClassifier(nn.Module):
|
11 |
+
def __init__(self, input_dim, num_classes):
|
12 |
+
super(DialectClassifier, self).__init__()
|
13 |
+
self.fc1 = nn.Linear(input_dim, 128)
|
14 |
+
self.fc2 = nn.Linear(128, 64)
|
15 |
+
self.fc3 = nn.Linear(64, num_classes)
|
16 |
+
self.relu = nn.ReLU()
|
17 |
+
|
18 |
+
def forward(self, x):
|
19 |
+
x = x.view(x.size(0), -1) # Flatten the input tensor
|
20 |
+
x = self.relu(self.fc1(x))
|
21 |
+
x = self.relu(self.fc2(x))
|
22 |
+
x = self.fc3(x)
|
23 |
+
return x
|
24 |
+
|
25 |
+
# Function to preprocess audio and extract features
|
26 |
+
def preprocess_audio(file_path, whisper_model, device):
|
27 |
+
def load_audio(file_path):
|
28 |
+
waveform, sample_rate = torchaudio.load(file_path)
|
29 |
+
if sample_rate != 16000:
|
30 |
+
waveform = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)(waveform)
|
31 |
+
|
32 |
+
# Convert to single channel (mono) if necessary
|
33 |
+
if waveform.shape[0] > 1:
|
34 |
+
waveform = torch.mean(waveform, dim=0, keepdim=True)
|
35 |
+
|
36 |
+
# Pad or trim audio to 30 seconds
|
37 |
+
desired_length = 16000 * 30 # 30 seconds at 16 kHz
|
38 |
+
current_length = waveform.shape[1]
|
39 |
+
|
40 |
+
if current_length < desired_length:
|
41 |
+
# Pad with zeros
|
42 |
+
padding = desired_length - current_length
|
43 |
+
waveform = torch.nn.functional.pad(waveform, (0, padding))
|
44 |
+
elif current_length > desired_length:
|
45 |
+
# Trim to desired length
|
46 |
+
waveform = waveform[:, :desired_length]
|
47 |
+
|
48 |
+
return waveform
|
49 |
+
|
50 |
+
audio = load_audio(file_path)
|
51 |
+
audio = whisper.pad_or_trim(audio.flatten())
|
52 |
+
mel = whisper.log_mel_spectrogram(audio).to_dense()
|
53 |
+
|
54 |
+
with torch.no_grad():
|
55 |
+
mel = mel.unsqueeze(0).to(device) # Add batch dimension and move to device
|
56 |
+
features = whisper_model.encoder(mel)
|
57 |
+
return features
|
58 |
+
|
59 |
+
repo_id = "dipankar53/assamese_dialect_classifier_model"
|
60 |
+
model_filename = "dialect_classifier_model.pth"
|
61 |
+
model_path = hf_hub_download(repo_id=repo_id, filename=model_filename)
|
62 |
+
|
63 |
+
label_to_idx = {"Darrangiya Accent": 0, "Kamrupiya Accent": 1, "Upper Assam": 2, "Nalbaria Accent": 3}
|
64 |
+
|
65 |
+
# Load Whisper model
|
66 |
+
whisper_model = whisper.load_model("medium")
|
67 |
+
|
68 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
69 |
+
|
70 |
+
# Load the trained model
|
71 |
+
num_classes = len(label_to_idx)
|
72 |
+
sample_input = torch.randn(1, 80, 3000).to(device)
|
73 |
+
with torch.no_grad():
|
74 |
+
sample_output = whisper_model.encoder(sample_input)
|
75 |
+
input_dim = sample_output.view(1, -1).shape[1] # Flatten and get dimension
|
76 |
+
|
77 |
+
model = DialectClassifier(input_dim, num_classes)
|
78 |
+
model.load_state_dict(torch.load(model_path, map_location=device))
|
79 |
+
model.to(device)
|
80 |
+
model.eval()
|
81 |
+
|
82 |
+
# Function to predict the dialect of a single audio file
|
83 |
+
def predict_dialect(audio_path):
|
84 |
+
try:
|
85 |
+
# Preprocess audio and extract features
|
86 |
+
features = preprocess_audio(audio_path, whisper_model, device)
|
87 |
+
features = features.view(1, -1) # Flatten features
|
88 |
+
|
89 |
+
# Perform prediction
|
90 |
+
with torch.no_grad():
|
91 |
+
outputs = model(features)
|
92 |
+
_, predicted = torch.max(outputs, 1)
|
93 |
+
|
94 |
+
# Map predicted index back to dialect label
|
95 |
+
idx_to_label = {idx: label for label, idx in label_to_idx.items()}
|
96 |
+
predicted_label = idx_to_label[predicted.item()]
|
97 |
+
|
98 |
+
return f"Predicted Dialect: {predicted_label}"
|
99 |
+
except Exception as e:
|
100 |
+
return f"Error: {str(e)}"
|
101 |
+
|
102 |
+
# Define Gradio interface
|
103 |
+
interface = gr.Interface(
|
104 |
+
fn=predict_dialect,
|
105 |
+
inputs=gr.Audio(sources=["upload", "microphone"], type="filepath"),
|
106 |
+
outputs="text",
|
107 |
+
title="Assamese Dialect Prediction",
|
108 |
+
description="Upload an Assamese audio file to predict its dialect.",
|
109 |
+
)
|
110 |
+
|
111 |
+
# Launch the interface
|
112 |
+
if __name__ == "__main__":
|
113 |
+
interface.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
torch
|
2 |
+
whisper
|
3 |
+
gradio
|
4 |
+
huggingface_hub
|
5 |
+
torchaudio
|