Primary competition visual

GIZ NLP Agricultural Keyword Spotter

Helping Uganda
$7 000 USD
Challenge completed almost 5 years ago
Classification
Automatic Speech Recognition
Natural Language Processing
737 joined
253 active
Starti
Sep 11, 20
Closei
Nov 29, 20
Reveali
Nov 29, 20
How to extract audio features using VGGish
Help · 6 Nov 2020, 20:07 · edited 7 minutes later · 2

I have tried following this link to extract audio features:

https://colab.research.google.com/drive/1TbX92UL9sYWbdwdGE0rJ9owmezB-Rl1C

but i unable to input raw audio files and get the 128 extracted features.

How can i extract the features

Discussion 2 answers

You can try using tensorflow model hub

import tensorflow as tf

import tensorflow_hub as hub

import numpy as np

import librosa

# Load the model.

vggmodel = hub.load('https://tfhub.dev/google/vggish/1')

def embedding_from_fn(fn):

x, sr = librosa.load(fn,sr=None)

x_16k = librosa.resample(x,sr,16000) #resample to 16KHz

embedding = np.array(vggmodel(x_16k))

return embedding

embedding = embedding_from_fn(df['fn'].values[0]).shape

6 Nov 2020, 20:17
Upvotes 0