speaker-verification-toolkit

speaker-verification-toolkit

This module contains some tools to make a simple speaker verification.

You can download it with PyPI:

$ pip install speaker-verification-toolkit

To import and use in your own projects:

import speaker_verification_toolkit.tools as svt

#   svt.some_function(...)

Usage

---

find_nearest_voice_data(voice_data_list, voice_sample)

Find the nearest voice data based on this voice sample. Could be used to make the naive Accept/Reject decision.

voice_data_list: a list containing all voices data from the dataset.

voice_sample: the voice sample reference.

returns: the index of the element from voice_data_list that represents the nearest voice data.

---

compute_distance(sample1, sample3)

Compute the distance between sample1 and sample2 using O(n) DTW algorithm

sample1: the mfcc data extracted from the audio signal 1.

sample2: the mfcc data extracted from the audio signal 2.

returns: Float number representing the minimum distance between sample1 and sample2.

---

extract_mfcc(signal_data, samplerate=16000, winlen=0.025, winstep=0.01)

Compute MFCC features from an audio signal

signal: the audio signal from which to compute features. Should be an N*1 array.

samplerate: the sample rate of the signal we are working with, in Hz.

winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds).

winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds).

returns: A numpy array of size (NUMFRAMES by numcep) containing features. Each row holds 1 feature vector.

---

extract_mfcc_from_wav_file(path, samplerate=16000, winlen=0.025, winstep=0.01)

Compute MFCC features from a wav file

path: the wav file path to be open.

samplerate: the wanted sample rate, in Hz. Default is 16000. If you want no resampling fill this argument with None.

winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds).

winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds).

returns: A numpy array of size (NUMFRAMES by numcep) containing features. Each row holds 1 feature vector.

---

rms_silence_filter(data, samplerate=16000, segment_length=None, threshold=0.001135)

Cut off silence parts from the signal audio data. Doesn't work with signals data affected by environment noise. You would consider apply a noise filter before using this silence filter or make sure that environment noise is small enough to be considered as silence.

data: the audio signal data

samplerate: if no segment_length is given, segment_length will be equals samplerate/100 (around 0.01 secs per segment).

segment_length: the number of frames per segment. I.e. for a sample rate SR, a segment length equals SR/100 will represent a chunk containing 0.01 seconds of audio.

threshold: the threshold value. Values less than or equal values will be cut off. The default value was defined at [1] (see the references).

returns: the param "data" without silence parts.

References

[1] - Muhammad Asadullah & Shibli Nisar, "A SILENCE REMOVAL AND ENDPOINT DETECTION APPROACH FOR SPEECH PROCESSING", National University of Computer and Emerging Sciences, Peshawar