app.py

import gradio as gr
import numpy as np
import pandas as pd
import joblib
import librosa
import plotly.express as px
import os

# Load the model and class mapping data once at the beginning
model = joblib.load('model.joblib')
class_mapping_data = pd.read_csv('data.csv')

# Preprocess the class mapping data for faster lookups
class_mapping_dict = dict(zip(class_mapping_data['encoded_label'],
                              class_mapping_data[['scientific_name', 'latitude', 'longitude', 'primary_label']].values))


# Define the feature extraction function
def extract_features(file_path):
    audio, _ = librosa.load(file_path)
    mfccs = librosa.feature.mfcc(y=audio, n_mfcc=40)
    return np.mean(mfccs.T, axis=0)


# Define the prediction function
def predict_bird(audio_file, sample_name):
    # Handle sample audio selection
    if sample_name in sample_audio_files:
        audio_file = sample_audio_files[sample_name]

    # Extract features from the audio file
    features = extract_features(audio_file)
    features = features.reshape(1, -1)

    # Predict the bird species
    prediction = model.predict(features)[0]

    # Retrieve bird information directly from the preprocessed dictionary
    bird_info = class_mapping_dict[prediction]
    predicted_bird = bird_info[0]

    # Calculate prediction confidence
    confidence = model.predict_proba(features)[0][prediction] * 100

    # Create a DataFrame for plotting
    tmp = pd.DataFrame([bird_info[1:]], columns=['latitude', 'longitude', 'primary_label'])

    # Create a scatter mapbox plot
    fig = px.scatter_mapbox(
        tmp,
        lat="latitude",
        lon="longitude",
        color="primary_label",
        zoom=10,
        title='Bird Recordings Location',
        mapbox_style="open-street-map"
    )
    fig.update_layout(margin={"r": 0, "t": 30, "l": 0, "b": 0})

    # Generate a spectrogram
    audio, sr = librosa.load(audio_file)
    spectrogram = librosa.feature.melspectrogram(y=audio, sr=sr)
    spectrogram_db = librosa.power_to_db(spectrogram, ref=np.max)
    fig_spectrogram = px.imshow(spectrogram_db, color_continuous_scale='viridis')
    fig_spectrogram.update_layout(title='Spectrogram', xaxis_title='Time', yaxis_title='Frequency')

    return predicted_bird, f"Confidence: {confidence:.2f}%", fig, fig_spectrogram


# Define sample audio files with full paths
sample_audio_files = {
    "Audio 1": os.path.join('sounds', 'asbfly.ogg'),
    "Audio 2": os.path.join('sounds', 'bkwsti.ogg'),
    "Audio 3": os.path.join('sounds', 'comros.ogg'),
}

# Create Gradio interface
iface = gr.Interface(
    fn=predict_bird,
    inputs=[
        gr.Audio(type="filepath", label="Upload Bird Sound"),
        gr.Dropdown(choices=list(sample_audio_files.keys()), label="Or Select a Sample", value=None)
    ],
    outputs=[
        gr.Textbox(label="Predicted Bird"),
        gr.Textbox(label="Prediction Confidence"),
        gr.Plot(label="Bird Recordings Location"),
        gr.Plot(label="Spectrogram"),
    ],
    title="Bird ID: Identify Bird Species from Audio Recordings",
    description="Upload an audio recording of a bird or select a sample to identify the species!",
)

# Launch the Gradio interface
iface.launch()