utils.py

import torch
import librosa
import numpy as np
import os


EMOTIONS = {
    1: "neural",
    2: "calm",
    3: "happy",
    4: "sad",
    5: "angry",
    6: "fear",
    7: "disgust",
    0: "suprise",
}

EMOTION_INTENSITY = {
    1: "normal",
    2: "strong",
}

class Utils():
    
    @staticmethod
    def get_devices() -> str:
        devices = "cpu"
        if torch.cuda.is_available():
            device = "cuda"
        elif torch.backends.mps.is_available():
            device = "mps"
        return device

    @staticmethod
    def extract_mfcc(file_path: str, sr: int) -> np.ndarray:
        audio, sample_rate = librosa.load(file_path, duration=3, offset=0.5, sr=sr)
        signal = np.zeros((int( sample_rate* 3)))
        signal[: len(audio)] = audio
        return signal


    @staticmethod
    def extract_mel_spectogram(data: np.ndarray, sr: int) -> np.ndarray:
        mel_spec = librosa.feature.melspectrogram(
            y=data,
            sr=sr,
            n_fft=1024,
            win_length=512,
            window="hamming",
            hop_length=256,
            n_mels=128,
            fmax=sr/ 2,
        )
        mel_spec_db = librosa.power_to_db(mel_spec, ref=np.max)
        return mel_spec_db

    @staticmethod
    def extract_features(path: str, sr: int):
        result = np.array([])
        audio_data, sample_rate = librosa.load(path, duration=2.5, offset=0.6, sr=sr)
        mfcc = extract_mfcc(audio_data, sample_rate)
        result = np.hstack((result, mfcc))
        mel_spec = extract_mel_spectogram(audio_data, sample_rate)
        result = np.hstack((result, mel_spec))
        return result


    @staticmethod
    def save_model(model):
        MODEL_PATH = os.path.join(os.getcwd(), "models")
        os.makedirs("models", exist_ok=True)
        torch.save(model.state_dict(), os.path.join(MODEL_PATH, "emotion_lstm.pt"))
        print("model saved")