audioperm 0.0.5

Audioperm, a python library for generating different permutations of audible segments from audio files.

Audioperm, a python library for generating different permutations of audible segments from audio files.

---

Audioperm

A python library for generating different permutations of audible segments from audio files.

pip install audioperm

Use:

• Silence Removal from Audio
• Audio / Speech augmentation
• Word segmentation
• Word level permutation generation
• Add new synthetic data for deep learning
• Speaker recognition, Speaker verification, Audio classification, Audio fingerprinting

Documentation: https://zabir-nabil.github.io/audioperm/

Source Code: https://github.com/zabir-nabil/audioperm

---

Word segmentation

from audioperm import AudioPerm
from audioperm.utils import save_audio

ap = AudioPerm("i_love_cats.m4a")
label = "i love cats"

words = ap.word_segments()
label_words = label.split()

for i, w in enumerate(words):
  save_audio(w, label_words[i] + ".wav")

cats.wav  i_love_cats.m4a  i.wav  love.wav

Word-level permutation

import numpy as np
from audioperm import AudioPerm
from audioperm.utils import save_audio

ap = AudioPerm("i_love_cats.m4a")
ap.word_segments(return_words=False)
perm_sentences = ap.permute(n_permutations = 5)

for i, s in enumerate(perm_sentences):
  save_audio(s, f"perm_{i}.wav")

cats.wav	   i.wav       perm_1.wav    perm_4.wav
i_love_cats.m4a    love.wav    perm_2.wav    perm_0.wav  
perm_3.wav

permutations on multiple files

from audioperm import read_audio, word_segments, permutations

ap = read_audio(["bangla_demo.wav", "i_love_cats.m4a"])
out = word_segments(ap)
perms = permutations(out, n_permutations = 5)

Fixed-length segments

• Generate fixed length audible segments (with permutation/augmentation)

from audioperm import fixed_len_segments
fixed_len_segments("bangla_demo.wav", return_segments = False, save_path = "fls_out", save = True, segment_size = 0.5)
out = fixed_len_segments("bangla_demo.wav", return_segments = True, max_segments = 5, permute = True, save = False, segment_size = 0.5)

Others

To run the code: Google Colab

Any contribution is welcome.

• Contributors
• Contribution guide

Tested with:

• python3.7
• python3.8

Internal audio representation:

• PCM 16
• float32

TO-DO:

• multi-channel audio
• augmentation
• multi-processing
• gpu-support