app.py

import torch
import torchaudio
from transformers import AutoFeatureExtractor, AutoModelForAudioClassification
import torch.nn.functional as F

# Load model
model_id = "Hatman/audio-emotion-detection"
extractor = AutoFeatureExtractor.from_pretrained(model_id)
model = AutoModelForAudioClassification.from_pretrained(model_id)

def classify_emotion(audio):
    waveform, sr = torchaudio.load(audio)

    # Mono
    if waveform.shape[0] > 1:
        waveform = waveform.mean(dim=0, keepdim=True)

    # Resample
    if sr != 16000:
        waveform = torchaudio.transforms.Resample(sr, 16000)(waveform)

    # Inference
    inputs = extractor(waveform.squeeze(0), sampling_rate=16000, return_tensors="pt")
    with torch.no_grad():
        logits = model(**inputs).logits
        probs = F.softmax(logits, dim=-1)[0]

    # Return scores as a dict
    return {model.config.id2label[i]: float(probs[i]) for i in range(len(probs))}

import gradio as gr
gr.Interface(fn=classify_emotion, 
             inputs=gr.Audio(type="filepath", label="Upload or record audio"), 
             outputs=gr.Label(num_top_classes=None), 
             title="Speech Emotion Detection").launch()