Spaces:
Sleeping
Sleeping
File size: 3,841 Bytes
857b1b2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 | import os
import sys
import time
import logging
import librosa
import numpy as np
import torch
from torch.nn import functional as F
from huggingface_hub import hf_hub_download
logger = logging.getLogger(__name__)
# HuggingFace repo
AVAILABLE_MODELS = {
"dual_emotion": "vyluong/emo_dual_classi"
}
emotion_labels = ['Angry', 'Anxiety', 'Happy', 'Sad', 'Neutral']
EMOTION_META = {
"Angry": {"emoji": "😡", "color": "#ff4d4f"},
"Anxiety": {"emoji": "😰", "color": "#faad14"},
"Happy": {"emoji": "😊", "color": "#52c41a"},
"Sad": {"emoji": "😢", "color": "#1890ff"},
"Neutral": {"emoji": "😐", "color": "#d9d9d9"},
}
class EmotionService:
_models = {}
emotion_labels = emotion_labels
meta = EMOTION_META
@classmethod
def load_dual_model(cls, repo_id, device):
logger.info(f"Downloading model from HF: {repo_id}")
model_file = hf_hub_download(
repo_id=repo_id,
filename="pytorch_model.bin"
)
model_code = hf_hub_download(
repo_id=repo_id,
filename="model.py"
)
# add model folder to python path
import importlib.util
spec = importlib.util.spec_from_file_location("hf_model", model_code)
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
Dual = module.Dual
model = Dual()
state_dict = torch.load(model_file, map_location=device)
model.load_state_dict(state_dict)
model.to(device)
model.eval()
logger.info("Emotion model loaded successfully")
return model
@classmethod
def get_model(cls, model_name="dual_emotion"):
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
if model_name in cls._models:
return cls._models[model_name]
repo_id = AVAILABLE_MODELS[model_name]
model = cls.load_dual_model(repo_id, device)
cls._models[model_name] = model
return model
@classmethod
def preload_model(cls):
logger.info("Preloading emotion model...")
cls.get_model()
logger.info("Emotion model ready")
# extract mfcc from segments
@staticmethod
def extract_mfcc_segment(
audio: np.ndarray,
sr: int,
start: float,
end: float,
duration: float = 5.0,
n_mfcc: int = 128,
n_fft: int = 2048,
hop_length: int = 512
):
start_sample = int(start * sr)
end_sample = int(end * sr)
segment = audio[start_sample:end_sample]
if len(segment) == 0:
return None
target_len = int(sr * duration)
if len(segment) < target_len:
segment = np.pad(segment,(0,target_len-len(segment)),mode="symmetric")
else:
segment = segment[:target_len]
mfcc = librosa.feature.mfcc(
y=segment,
sr=sr,
n_mfcc=n_mfcc,
n_fft=n_fft,
hop_length=hop_length
)
return mfcc
@classmethod
def predict_from_mfcc(cls, mfcc):
model = cls.get_model()
tensor = torch.from_numpy(mfcc).unsqueeze(0).unsqueeze(0).float()
device = next(model.parameters()).device
tensor = tensor.to(device)
with torch.no_grad():
output = model(tensor)
probs = F.softmax(output.squeeze(), dim=0).cpu().numpy()
label = cls.emotion_labels[np.argmax(probs)]
return label
# predict from segments
@classmethod
def predict_segment(cls, audio, sr, start, end):
mfcc = cls.extract_mfcc_segment(audio, sr, start, end)
if mfcc is None:
return "Neutral"
return cls.predict_from_mfcc(mfcc) |