app.py

import gradio as gr
import numpy as np
import tensorflow as tf
import librosa
from pydub import AudioSegment
import pickle
import gdown
import os

# Download the model from Google Drive
model_url = "https://drive.google.com/uc?id=1Z7Of2hr48B3c1mb_j0WV8y6JaTkFvrUe"
output = "best_model.keras"
if not os.path.exists(output):
    gdown.download(model_url, output, quiet=False)

# Load the species names
with open('species_names_resnet.pkl', 'rb') as f:
    species_names_resnet = pickle.load(f)

# Load the trained model
model = tf.keras.models.load_model('best_model.keras')

# Define preprocessing functions
def pad_audio(segment, target_duration_ms):
    if len(segment) < target_duration_ms:
        padding = AudioSegment.silent(duration=target_duration_ms - len(segment))
        segment = segment + padding
    return segment

def preprocess_audio(audio_path, target_duration_ms=3000, sr=22050, n_mels=96, fmax=8000):
    segment = AudioSegment.from_file(audio_path)
    segment = pad_audio(segment, target_duration_ms)
    
    samples = np.array(segment.get_array_of_samples(), dtype=np.float32) / 32768.0
    mel_spectrogram = librosa.feature.melspectrogram(y=samples, sr=sr, n_mels=n_mels, fmax=fmax)
    log_mel_spectrogram = librosa.power_to_db(mel_spectrogram, ref=np.max)
    
    if log_mel_spectrogram.shape[1] < 130:
        log_mel_spectrogram = np.pad(log_mel_spectrogram, ((0, 0), (0, 130 - log_mel_spectrogram.shape[1])), mode='constant')
    elif log_mel_spectrogram.shape[1] > 130:
        log_mel_spectrogram = log_mel_spectrogram[:, :130]
    
    return log_mel_spectrogram.T  # Transpose to [130, 96]

def predict(audio_file):
    features = preprocess_audio(audio_file)
    features = features[np.newaxis, ..., np.newaxis]  # Add batch and channel dimensions
    features = np.repeat(features, 3, axis=-1)  # Repeat to create 3 channels
    features = np.transpose(features, (0, 2, 1, 3))  # Transpose to match the expected shape (None, 96, 130, 3)
    
    y_pred_prob = model.predict(features)
    y_pred_label = np.argmax(y_pred_prob, axis=1)
    
    return species_names_resnet[y_pred_label[0]]


# Create Gradio interface
interface = gr.Interface(
    fn=predict,
    inputs=gr.Audio(type="filepath"),
    outputs="text",
    title="Bird Species Prediction",
    description="Upload an audio file to predict the bird species."
)


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