Spaces:
Sleeping
Sleeping
Upload 3 files
Browse files- app (1).py +103 -0
- mymodel_SER_LSTM_RAVDESS (1).h5 +3 -0
- requirements.txt +12 -0
app (1).py
ADDED
@@ -0,0 +1,103 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import numpy as np
|
3 |
+
import librosa
|
4 |
+
import requests
|
5 |
+
from io import BytesIO
|
6 |
+
from PIL import Image
|
7 |
+
import os
|
8 |
+
from tensorflow.keras.models import load_model
|
9 |
+
from faster_whisper import WhisperModel
|
10 |
+
|
11 |
+
# Load the emotion prediction model
|
12 |
+
def load_emotion_model(model_path):
|
13 |
+
try:
|
14 |
+
model = load_model(model_path)
|
15 |
+
return model
|
16 |
+
except Exception as e:
|
17 |
+
print("Error loading emotion prediction model:", e)
|
18 |
+
return None
|
19 |
+
|
20 |
+
model_path = 'mymodel_SER_LSTM_RAVDESS.h5'
|
21 |
+
model = load_emotion_model(model_path)
|
22 |
+
|
23 |
+
# Initialize WhisperModel
|
24 |
+
model_size = "small"
|
25 |
+
model2 = WhisperModel(model_size, device="cpu", compute_type="int8")
|
26 |
+
|
27 |
+
# Function to transcribe audio
|
28 |
+
def transcribe(wav_filepath):
|
29 |
+
segments, _ = model2.transcribe(wav_filepath, beam_size=5)
|
30 |
+
return "".join([segment.text for segment in segments])
|
31 |
+
|
32 |
+
# Function to extract MFCC features from audio
|
33 |
+
def extract_mfcc(wav_file_name):
|
34 |
+
try:
|
35 |
+
y, sr = librosa.load(wav_file_name)
|
36 |
+
mfccs = np.mean(librosa.feature.mfcc(y=y, sr=sr, n_mfcc=40).T, axis=0)
|
37 |
+
return mfccs
|
38 |
+
except Exception as e:
|
39 |
+
print("Error extracting MFCC features:", e)
|
40 |
+
return None
|
41 |
+
|
42 |
+
# Emotions dictionary
|
43 |
+
emotions = {1: 'neutral', 2: 'calm', 3: 'happy', 4: 'sad', 5: 'angry', 6: 'fearful', 7: 'disgust', 8: 'surprised'}
|
44 |
+
|
45 |
+
# Function to predict emotion from audio
|
46 |
+
def predict_emotion_from_audio(wav_filepath):
|
47 |
+
try:
|
48 |
+
test_point = extract_mfcc(wav_filepath)
|
49 |
+
if test_point is not None:
|
50 |
+
test_point = np.reshape(test_point, newshape=(1, 40, 1))
|
51 |
+
predictions = model.predict(test_point)
|
52 |
+
predicted_emotion_label = np.argmax(predictions[0]) + 1
|
53 |
+
return emotions[predicted_emotion_label]
|
54 |
+
else:
|
55 |
+
return "Error: Unable to extract features"
|
56 |
+
except Exception as e:
|
57 |
+
print("Error predicting emotion:", e)
|
58 |
+
return None
|
59 |
+
|
60 |
+
api_key = os.getenv("DeepAI_api_key")
|
61 |
+
|
62 |
+
# Function to generate an image using DeepAI Text to Image API
|
63 |
+
def generate_image(api_key, text):
|
64 |
+
url = "https://api.deepai.org/api/text2img"
|
65 |
+
headers = {'api-key': api_key}
|
66 |
+
response = requests.post(
|
67 |
+
url,
|
68 |
+
data={'text': text},
|
69 |
+
headers=headers
|
70 |
+
)
|
71 |
+
response_data = response.json()
|
72 |
+
if 'output_url' in response_data:
|
73 |
+
image_url = response_data['output_url']
|
74 |
+
image_response = requests.get(image_url)
|
75 |
+
image = Image.open(BytesIO(image_response.content))
|
76 |
+
return image
|
77 |
+
else:
|
78 |
+
return None
|
79 |
+
|
80 |
+
|
81 |
+
# Function to get predictions
|
82 |
+
def get_predictions(audio_input):
|
83 |
+
emotion_prediction = predict_emotion_from_audio(audio_input)
|
84 |
+
transcribed_text = transcribe(audio_input)
|
85 |
+
texto_imagen = emotion_prediction + transcribed_text
|
86 |
+
image = generate_image(api_key, texto_imagen)
|
87 |
+
return emotion_prediction, transcribed_text, image
|
88 |
+
|
89 |
+
# Create the Gradio interface
|
90 |
+
interface = gr.Interface(
|
91 |
+
fn=get_predictions,
|
92 |
+
inputs=gr.Audio(label="Input Audio", type="filepath"),
|
93 |
+
outputs=[
|
94 |
+
gr.Label("Acoustic Prediction", label="Acoustic Prediction"),
|
95 |
+
gr.Label("Transcribed Text", label="Transcribed Text"),
|
96 |
+
gr.Image(type='pil', label="Generated Image")
|
97 |
+
],
|
98 |
+
title="Affective Virtual Environments",
|
99 |
+
description="Create an AVE using your voice."
|
100 |
+
)
|
101 |
+
|
102 |
+
|
103 |
+
interface.launch()
|
mymodel_SER_LSTM_RAVDESS (1).h5
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:5ad725c49cec0f25f17e1c798f35d3b3e486ffdf2cf97497f2beb99805dc6c8f
|
3 |
+
size 976728
|
requirements.txt
ADDED
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
kaleido
|
2 |
+
numpy
|
3 |
+
tensorflow==2.12.0
|
4 |
+
gradio
|
5 |
+
transformers
|
6 |
+
tf-keras
|
7 |
+
librosa
|
8 |
+
vaderSentiment
|
9 |
+
requests
|
10 |
+
torch
|
11 |
+
sentencepiece
|
12 |
+
faster_whisper
|