app.py

import gradio as gr
import torch
import torchaudio
import numpy as np
from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor, AutoConfig
import matplotlib.pyplot as plt

# =========================
# CONFIG
# =========================
MODEL_NAME = "Hatman/audio-emotion-detection"
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

# =========================
# LOAD MODEL & FEATURE EXTRACTOR
# =========================
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(MODEL_NAME)
model = Wav2Vec2ForSequenceClassification.from_pretrained(MODEL_NAME).to(DEVICE)
model.eval()

# Use the model’s label mapping directly
config = AutoConfig.from_pretrained(MODEL_NAME)
LABELS = [config.id2label[i] for i in range(len(config.id2label))]

# Map some emojis to each emotion for fun UI
EMOJIS = {
    "Angry": "😡",
    "Disgusted": "🤢",
    "Fearful": "😨",
    "Happy": "😄",
    "Neutral": "😐",
    "Sad": "😢",
    "Surprised": "😲"
}

# =========================
# PREDICTION FUNCTION
# =========================
def predict(audio):
    try:
        if audio is None:
            return {"Error": "No audio provided"}, None

        sr, data = audio
        data = np.array(data, dtype=np.float32)

        # Stereo -> Mono
        if len(data.shape) > 1:
            data = np.mean(data, axis=1)

        # Resample to 16kHz
        if sr != 16000:
            data = torchaudio.functional.resample(torch.tensor(data), sr, 16000).numpy()
            sr = 16000

        # Normalize for Wav2Vec2
        data = data / 32768.0

        # Feature extraction
        inputs = feature_extractor(
            data,
            sampling_rate=sr,
            return_tensors="pt",
            padding=True
        )

        # Move tensors to device
        for k in inputs:
            inputs[k] = inputs[k].to(DEVICE)

        # Forward pass
        with torch.no_grad():
            logits = model(**inputs).logits
            probs = torch.nn.functional.softmax(logits, dim=-1)[0].cpu().numpy()

        # Format top 3 results with emojis
        top_idx = np.argsort(probs)[::-1][:3]
        result = {f"{LABELS[i]} {EMOJIS.get(LABELS[i], '')}": round(float(probs[i]), 4) for i in top_idx}

        # Generate waveform plot
        fig, ax = plt.subplots(figsize=(6,2))
        ax.plot(data, color='purple')
        ax.set_title("Audio Waveform")
        ax.set_xlabel("Samples")
        ax.set_ylabel("Amplitude")
        ax.set_xticks([])
        ax.set_yticks([])
        plt.tight_layout()

        return result, fig

    except Exception as e:
        return {"Error": str(e)}, None

# =========================
# GRADIO APP
# =========================
demo = gr.Interface(
    fn=predict,
    inputs=gr.Audio(sources=["upload", "microphone"], type="numpy", label="🎤 Upload or Record Audio"),
    outputs=[gr.Label(num_top_classes=3), gr.Plot()],
    title="Audio Emotion Detection 🎧",
    description=(
        "Fine-tuned Wav2Vec2 model (`Hatman/audio-emotion-detection`) "
        "for emotion recognition from voice. "
        "Detects: Angry, Disgusted, Fearful, Happy, Neutral, Sad, Surprised. "
        "Audio is auto-resampled to 16kHz."
    ),
    allow_flagging="never",
)

# =========================
# LAUNCH
# =========================
if __name__ == "__main__":
    demo.launch()