app.py

# app.py

import gradio as gr
import torch
import numpy as np
import librosa
from transformers import Wav2Vec2FeatureExtractor, Wav2Vec2ForSequenceClassification

# ------------------------------------------------
# 1. Load base Wav2Vec2 model + classification head
# ------------------------------------------------
model_name = "facebook/wav2vec2-base-960h"

# Specify num_labels=8 to create a random classification head on top.
model = Wav2Vec2ForSequenceClassification.from_pretrained(
    model_name,
    num_labels=8
)
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_name)

model.eval()

# ------------------------------------------------
# 2. Define inference function
# ------------------------------------------------
def classify_accuracy(audio):
    """
    Receives a tuple (sample_rate, data) from Gradio when type='numpy'.
    Resamples if needed, runs a forward pass, and returns a 'level'.
    """
    if audio is None:
        return "No audio provided."

    sample_rate, data = audio

    # Ensure data is a NumPy array.
    if not isinstance(data, np.ndarray):
        data = np.array(data)

    # Resample to 16kHz if needed.
    target_sr = 16000
    if sample_rate != target_sr:
        data = librosa.resample(data, orig_sr=sample_rate, target_sr=target_sr)
        sample_rate = target_sr

    # Extract features from the audio data.
    inputs = feature_extractor(
        data,
        sampling_rate=sample_rate,
        return_tensors="pt",
        padding=True
    )

    # Run model inference.
    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        predicted_id = torch.argmax(logits, dim=-1).item()

    # Map predicted id (0..7) to the final level (3..10).
    accuracy_level = predicted_id + 3
    return f"Predicted Accuracy Level: {accuracy_level}"

# ------------------------------------------------
# 3. Build Gradio interface
# ------------------------------------------------
title = "Speech Accuracy Classifier (Base Wav2Vec2)"
description = (
    "Record audio using your microphone or upload an audio file (left). "
    "The model (not fine-tuned) will classify the audio into an accuracy level (right)."
)

# Using source="microphone" allows for direct recording, while recent versions also enable file upload.
demo = gr.Interface(
    fn=classify_accuracy,
    inputs=gr.Audio(source="microphone", type="numpy", label="Record/Upload Audio"),
    outputs=gr.Textbox(label="Classification Result"),
    title=title,
    description=description,
    allow_flagging="never"
)

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