Runs Google Media Pipe Yamnet Audio Classification on MP3 File

GoogleMediaPipeAudioClassification.py

import numpy as np
from pprint import pprint
import librosa
import tensorflow as tf

# https://storage.googleapis.com/mediapipe-models/audio_classifier/yamnet/float32/latest/yamnet.tflite
# https://github.com/tensorflow/models/blob/master/research/audioset/yamnet/yamnet.py
# https://research.google.com/audioset/ontology/index.html
# https://storage.googleapis.com/mediapipe-tasks/audio_classifier/yamnet_label_list.txt

# https://github.com/tensorflow/models/blob/master/research/audioset/yamnet/params.py#L25
sample_rate = 16000.0
stft_window_seconds = 0.025
stft_hop_seconds = 0.010
mel_bands = 64
mel_min_hz = 125.0
mel_max_hz = 7500.0
log_offset = 0.001
# patch_window_seconds = 0.96 # or 0.975 ?
patch_window_seconds = 0.975
patch_hop_seconds = 0.48

interpreter = tf.lite.Interpreter( model_path='./yamnet.tflite' )
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()

def read_text( file_path ):
	with open( file_path ) as f:
		return f.read().splitlines()

def process_audio_segment( segment ):
	waveform = segment.astype( np.float32 )
	interpreter.set_tensor( input_details[ 0 ][ "index" ] , waveform )
	interpreter.invoke()
	return interpreter.get_tensor( output_details[ 0 ][ "index" ] ).flatten()

if __name__ == "__main__":

	model_labels = read_text( "./yamnet_label_list.txt" )

	# classify
	audio_path = './Tradition.mp3'
	audio_data, sr = librosa.load( audio_path , sr=sample_rate )

	samples_per_patch = ( patch_window_seconds * sample_rate )
	samples_per_patch_50_percent_overlap = int( samples_per_patch / 2 )

	results = []
	duration_in_samples = len( audio_data )
	range_end = int( duration_in_samples - ( samples_per_patch + 1 ) )
	for start in range( 0 , range_end , samples_per_patch_50_percent_overlap ):  # 50% overlap
		segment = audio_data[ start: ( start + int( samples_per_patch ) ) ]
		probabilities = process_audio_segment( segment )
		results.append( probabilities )

	total_results = len( results )
	for i , result in enumerate( results ):
		print( f"\nSection [{i+1}] of {total_results}" )
		labeled_probabilities = list( zip( model_labels , result ) )
		labeled_probabilities = [ pair for pair in labeled_probabilities if pair[ 1 ] > 0 ]
		labeled_probabilities.sort( key=lambda x: x[ 1 ] , reverse=True )
		for label, probability in labeled_probabilities[ : 19 ]:
			print( f"{label}: {probability}" )