ML-with-Rajibul
commited on
Commit
•
9910ecc
1
Parent(s):
5f93bf1
Upload 5 files
Browse files- .gitattributes +1 -0
- MT.py +34 -0
- SER.py +115 -0
- X_train.pkl +3 -0
- Y_train.pkl +3 -0
- speech-emotion-recognition.hdf5 +3 -0
.gitattributes
CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
36 |
+
speech-emotion-recognition.hdf5 filter=lfs diff=lfs merge=lfs -text
|
MT.py
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import random
|
2 |
+
import spotipy
|
3 |
+
from spotipy.oauth2 import SpotifyClientCredentials
|
4 |
+
|
5 |
+
# Authenticate with Spotify API
|
6 |
+
client_id = '471e06ff0a13445095909029b18c265c'
|
7 |
+
client_secret = 'c0f56895d29f434cbeac4309d0b42d05'
|
8 |
+
client_credentials_manager = SpotifyClientCredentials(client_id=client_id, client_secret=client_secret)
|
9 |
+
sp = spotipy.Spotify(client_credentials_manager=client_credentials_manager)
|
10 |
+
|
11 |
+
def search_song_by_emotion(emotion):
|
12 |
+
# Define a mapping of emotions to search keywords
|
13 |
+
emotion_keywords = {
|
14 |
+
"neutral": ["raga des sarangi", "raga malkauns", "raga bhairav", "raga rageshri"],
|
15 |
+
"surprise": ["raag hameer", "raag kedar", "raga puriya"],
|
16 |
+
"fear": ["raag bilahari", "raag purvi", "raag shudh kalyan", "raag miya ki malhar"],
|
17 |
+
"sad": ["raag yaman sitar", "raga hameer","raga shyam kalyan"],
|
18 |
+
"angry": ["raag jaijaiwanti", "raag bhairavi", "raga puriya", "raag kafi"],
|
19 |
+
"happy": ["raga hamsadhwani sarod", "raga khamaj", "raga bhupali", "raga bahar"],
|
20 |
+
"disgust": ["raga khamaj", "raga bilaskhani todi", "raga shudh kalyan", "raga puriya"]
|
21 |
+
}
|
22 |
+
|
23 |
+
# Search for playlists based on the emotion keywords
|
24 |
+
keywords = emotion_keywords.get(emotion.lower(), [])
|
25 |
+
if keywords:
|
26 |
+
keywords = random.choice(keywords)
|
27 |
+
results = sp.search(q=f"track:{keywords}", type="track", limit=1)
|
28 |
+
tracks = results["tracks"]["items"]
|
29 |
+
|
30 |
+
# Extract song previews from the playlists
|
31 |
+
if tracks:
|
32 |
+
preview_url = tracks[0]["preview_url"]
|
33 |
+
return preview_url
|
34 |
+
return None
|
SER.py
ADDED
@@ -0,0 +1,115 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
import numpy as np
|
3 |
+
|
4 |
+
import librosa
|
5 |
+
|
6 |
+
import sklearn
|
7 |
+
from sklearn.preprocessing import StandardScaler, OneHotEncoder
|
8 |
+
from sklearn.model_selection import train_test_split
|
9 |
+
|
10 |
+
import tensorflow as tf
|
11 |
+
from keras.models import load_model
|
12 |
+
|
13 |
+
import pickle
|
14 |
+
|
15 |
+
sample_rate = 22050
|
16 |
+
|
17 |
+
def noise(data):
|
18 |
+
noise_value = 0.015 * np.random.uniform() * np.amax(data)
|
19 |
+
data = data + noise_value * np.random.normal(size=data.shape[0])
|
20 |
+
return data
|
21 |
+
|
22 |
+
def stretch(data, rate=0.8):
|
23 |
+
return librosa.effects.time_stretch(data, rate=rate)
|
24 |
+
|
25 |
+
def shift(data):
|
26 |
+
shift_range = int(np.random.uniform(low=-5, high=5) * 1000)
|
27 |
+
return np.roll(data, shift_range)
|
28 |
+
|
29 |
+
def pitch(data,sampling_rate,pitch_factor=0.7):
|
30 |
+
return librosa.effects.pitch_shift(data,sr=sampling_rate, n_steps=pitch_factor)
|
31 |
+
|
32 |
+
def extract_process(data):
|
33 |
+
|
34 |
+
sample_rate = 22050
|
35 |
+
output_result = np.array([])
|
36 |
+
mean_zero = np.mean(librosa.feature.zero_crossing_rate(y=data).T,axis=0)
|
37 |
+
output_result = np.hstack((output_result,mean_zero))
|
38 |
+
|
39 |
+
stft_out = np.abs(librosa.stft(data))
|
40 |
+
chroma_stft = np.mean(librosa.feature.chroma_stft(S=stft_out,sr=sample_rate).T,axis=0)
|
41 |
+
output_result = np.hstack((output_result,chroma_stft))
|
42 |
+
|
43 |
+
mfcc_out = np.mean(librosa.feature.mfcc(y=data,sr=sample_rate).T,axis=0)
|
44 |
+
output_result = np.hstack((output_result,mfcc_out))
|
45 |
+
|
46 |
+
root_mean_out = np.mean(librosa.feature.rms(y=data).T,axis=0)
|
47 |
+
output_result = np.hstack((output_result,root_mean_out))
|
48 |
+
|
49 |
+
mel_spectogram = np.mean(librosa.feature.melspectrogram(y=data,sr=sample_rate).T,axis=0)
|
50 |
+
output_result = np.hstack((output_result,mel_spectogram))
|
51 |
+
|
52 |
+
return output_result
|
53 |
+
|
54 |
+
def export_process(path):
|
55 |
+
|
56 |
+
data,sample_rate = librosa.load(path,duration=2.5,offset=1)
|
57 |
+
|
58 |
+
output_1 = extract_process(data)
|
59 |
+
result = np.array(output_1)
|
60 |
+
|
61 |
+
noise_out = noise(data)
|
62 |
+
output_2 = extract_process(noise_out)
|
63 |
+
result = np.vstack((result,output_2))
|
64 |
+
|
65 |
+
new_out = stretch(data)
|
66 |
+
strectch_pitch = pitch(new_out,sample_rate)
|
67 |
+
output_3 = extract_process(strectch_pitch)
|
68 |
+
result = np.vstack((result,output_3))
|
69 |
+
|
70 |
+
return result
|
71 |
+
|
72 |
+
# Load X_train from Google Drive
|
73 |
+
with open('X_train.pkl', 'rb') as f:
|
74 |
+
X_train = pickle.load(f)
|
75 |
+
|
76 |
+
# Load X_train from Google Drive
|
77 |
+
with open('Y_train.pkl', 'rb') as f:
|
78 |
+
Y_train = pickle.load(f)
|
79 |
+
|
80 |
+
Features = pd.DataFrame(X_train)
|
81 |
+
Features['labels'] = Y_train
|
82 |
+
|
83 |
+
X = Features.iloc[: ,:-1].values
|
84 |
+
Y = Features['labels'].values
|
85 |
+
|
86 |
+
encoder_label = OneHotEncoder()
|
87 |
+
Y = encoder_label.fit_transform(np.array(Y).reshape(-1,1)).toarray()
|
88 |
+
|
89 |
+
x_train, x_test, y_train, y_test = train_test_split(X, Y, train_size=0.9, random_state=42, shuffle=True)
|
90 |
+
|
91 |
+
scaler_data = StandardScaler()
|
92 |
+
x_train = scaler_data.fit_transform(x_train)
|
93 |
+
x_test = scaler_data.transform(x_test)
|
94 |
+
|
95 |
+
def preprocess_audio(audio):
|
96 |
+
#data, sample_rate = librosa.load(audio, duration=2.5, offset=0.6)
|
97 |
+
features = export_process(audio)
|
98 |
+
features = scaler_data.transform(features)
|
99 |
+
return np.expand_dims(features, axis=2)
|
100 |
+
|
101 |
+
# Function to predict emotion from preprocessed audio
|
102 |
+
def predict_emotion(preprocessed_audio):
|
103 |
+
model = load_model('speech-emotion-recognition.hdf5')
|
104 |
+
prediction = model.predict(preprocessed_audio)
|
105 |
+
predicted_emotion = encoder_label.inverse_transform(prediction)
|
106 |
+
return predicted_emotion[0]
|
107 |
+
|
108 |
+
# Live emotion recognition
|
109 |
+
def live_emotion_recognition(audio_path):
|
110 |
+
# Preprocess live audio
|
111 |
+
preprocessed_audio = preprocess_audio(audio_path)
|
112 |
+
# Predict emotion
|
113 |
+
predicted_emotion = predict_emotion(preprocessed_audio)
|
114 |
+
#print("Predicted Emotion:", predicted_emotion)
|
115 |
+
return predicted_emotion[0]
|
X_train.pkl
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3ca0b38da847556205b8092899f4472153bf65fd3af95055d5df4c51720c44e2
|
3 |
+
size 11165240
|
Y_train.pkl
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ade5a4397190642531e4d909cb18b54aa5f12cae4b587483f8190576ac01c8b0
|
3 |
+
size 48108
|
speech-emotion-recognition.hdf5
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d9efa37a959fbc465d3a96912383c28cef9a35c3d0cb2065abda86f58c13ee32
|
3 |
+
size 6747280
|