File size: 3,154 Bytes
7651694 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 |
import gradio as gr
import numpy as np
import librosa
from PIL import Image
import tempfile
import os
from app.preprocess import preprocess_audio
from app.model import predict
from collections import Counter, defaultdict
# Process Image Input
def process_image_input(img):
"""Classify a spectrogram image directly using model.predict"""
label, confidence, probs = predict(img)
return label, round(confidence, 3), probs
# Process Audio Input
def process_audio_input(audio_file):
# Save uploaded audio temporarily
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
tmp.write(audio_file)
tmp_path = tmp.name
# Preprocess → mel-spectrogram chunks (list of PIL images)
imgs = preprocess_audio(tmp_path)
os.remove(tmp_path)
# Predict on each chunk
all_preds = []
all_confs = []
all_probs = []
for img in imgs:
label, conf, probs = predict(img)
all_preds.append(label)
all_confs.append(conf)
all_probs.append(probs)
# Majority Vote
counter = Counter(all_preds)
max_count = max(counter.values())
candidates = [k for k, v in counter.items() if v == max_count]
if len(candidates) == 1:
final_label = candidates[0]
else:
conf_sums = defaultdict(float)
for i, label in enumerate(all_preds):
if label in candidates:
conf_sums[label] += all_confs[i]
final_label = max(conf_sums, key=conf_sums.get)
final_conf = float(np.mean([all_confs[i] for i, l in enumerate(all_preds) if l == final_label]))
return final_label, round(final_conf, 3), all_preds, [round(c, 3) for c in all_confs]
# MAIN GRADIO CLASSIFICATION PIPELINE (AUDIO OR IMAGE)
def classify(audio, image):
# If image is provided → classify image
if image is not None:
label, conf, probs = process_image_input(image)
return {
"Final Label": label,
"Confidence": conf,
"Details": probs
}
# If audio is provided → preprocess audio → classify
if audio is not None:
label, conf, all_preds, all_confs = process_audio_input(audio)
return {
"Final Label": label,
"Confidence": conf,
"All Chunk Labels": all_preds,
"All Chunk Confidences": all_confs
}
# Nothing provided
return "Please upload an audio file OR a spectrogram image."
# GRADIO UI
interface = gr.Interface(
fn=classify,
inputs=[
gr.Audio(type="bytes", label="Upload Audio (WAV/MP3)"),
gr.Image(type="pil", label="Upload Spectrogram Image")
],
outputs=gr.JSON(label="Prediction Results"),
title="General Audio Classifier (Audio + Spectrogram Support)",
description=(
"Upload a raw audio file OR a spectrogram image.\n"
"The app automatically detects the input type:\n"
"• If audio → the model preprocesses it into mel spectrogram chunks.\n"
"• If spectrogram → the model classifies it directly.\n"
"Built using CNN + Mel-Spectrogram + Gradio."
),
)
interface.launch() |