src/models/model-manager.py

from transformers import (
    WhisperProcessor, WhisperForConditionalGeneration,
    AutoModelForSequenceClassification, AutoTokenizer
)
import torch

class ModelManager:
    def __init__(self):
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        self.models = {}
        self.tokenizers = {}
        self.processors = {}
        
    def load_models(self):
        # Load Whisper for speech recognition
        self.processors['whisper'] = WhisperProcessor.from_pretrained("openai/whisper-base")
        self.models['whisper'] = WhisperForConditionalGeneration.from_pretrained(
            "openai/whisper-base"
        ).to(self.device)
        
        # Load EmoBERTa for emotion detection
        self.tokenizers['emotion'] = AutoTokenizer.from_pretrained("arpanghoshal/EmoRoBERTa")
        self.models['emotion'] = AutoModelForSequenceClassification.from_pretrained(
            "arpanghoshal/EmoRoBERTa"
        ).to(self.device)
        
        # Load ClinicalBERT for analysis
        self.tokenizers['clinical'] = AutoTokenizer.from_pretrained(
            "emilyalsentzer/Bio_ClinicalBERT"
        )
        self.models['clinical'] = AutoModelForSequenceClassification.from_pretrained(
            "emilyalsentzer/Bio_ClinicalBERT"
        ).to(self.device)
        
    def transcribe(self, audio_input):
        inputs = self.processors['whisper'](
            audio_input, 
            return_tensors="pt"
        ).input_features.to(self.device)
        
        generated_ids = self.models['whisper'].generate(inputs)
        transcription = self.processors['whisper'].batch_decode(
            generated_ids, 
            skip_special_tokens=True
        )[0]
        return transcription
        
    def analyze_emotions(self, text):
        inputs = self.tokenizers['emotion'](
            text,
            return_tensors="pt",
            padding=True,
            truncation=True,
            max_length=512
        ).to(self.device)
        
        outputs = self.models['emotion'](**inputs)
        probs = torch.nn.functional.softmax(outputs.logits, dim=-1)
        
        emotions = ['anger', 'fear', 'joy', 'love', 'sadness', 'surprise']
        return {emotion: float(prob) for emotion, prob in zip(emotions, probs[0])}
        
    def analyze_mental_health(self, text):
        inputs = self.tokenizers['clinical'](
            text,
            return_tensors="pt",
            padding=True,
            truncation=True,
            max_length=512
        ).to(self.device)
        
        outputs = self.models['clinical'](**inputs)
        scores = torch.sigmoid(outputs.logits)
        
        return {
            'depression_risk': float(scores[0][0]),
            'anxiety_risk': float(scores[0][1]),
            'stress_level': float(scores[0][2])
        }