| # emotion_detector.py | |
| import os | |
| import torchaudio | |
| import tempfile | |
| from speechbrain.inference.classifiers import AudioClassifier | |
| # Set HF_HOME to writable temp directory | |
| os.environ["HF_HOME"] = os.path.join(tempfile.gettempdir(), "hf_cache") | |
| # Also set Hugging Face cache dir explicitly (optional but helps) | |
| os.environ["HF_DATASETS_CACHE"] = os.environ["HF_HOME"] | |
| os.environ["TRANSFORMERS_CACHE"] = os.environ["HF_HOME"] | |
| os.environ["HUGGINGFACE_HUB_CACHE"] = os.environ["HF_HOME"] | |
| # Use a temp directory for SpeechBrain model saving | |
| model_dir = os.path.join(tempfile.gettempdir(), "emotion_model") | |
| # Load pretrained model | |
| classifier = AudioClassifier.from_hparams( | |
| source="speechbrain/emotion-recognition-wav2vec2-IEMOCAP", | |
| savedir=model_dir | |
| ) | |
| EMOTION_EMOJIS = { | |
| "angry": "😠", | |
| "happy": "😄", | |
| "neutral": "😐", | |
| "sad": "😢", | |
| "fearful": "😨" | |
| } | |
| def detect_emotion(audio_path): | |
| signal, fs = torchaudio.load(audio_path) | |
| prediction = classifier.classify_file(audio_path) | |
| emotion = prediction[3] # Predicted label | |
| emoji = EMOTION_EMOJIS.get(emotion.lower(), "❓") | |
| return emotion, emoji | |