We use cookies on this website to distinguish you from other users. We use this data to improve our content experience and for targeted advertising. By continuing to use this website you consent to our use of cookies. For more information, please see our Cookie Policy.

spafe.fbanks.mel_fbanks#

Description : Mel filter banks implementation.
Copyright (c) 2019-2024 Ayoub Malek. This source code is licensed under the terms of the BSD 3-Clause License. For a copy, see <https://github.com/SuperKogito/spafe/blob/master/LICENSE>.

spafe.fbanks.mel_fbanks.mel_filter_banks_helper(nfilts: int = 24, nfft: int = 512, fs: int = 16000, low_freq: float = 0, high_freq: Optional[float] = None, scale: typing_extensions.Literal[ascendant, descendant, constant] = 'constant', fb_type='mel', conversion_approach: typing_extensions.Literal[Oshaghnessy, Lindsay] = 'Oshaghnessy')[source]#

Compute Mel-filter banks.The filters are stored in the rows, the columns correspond to fft bins.

Parameters

nfilts (int) – the number of filters in the filter bank. (Default 20).
nfft (int) – the FFT size. (Default is 512).
fs (int) – sample rate/ sampling frequency of the signal. (Default 16000 Hz).
low_freq (float) – lowest band edge of mel filters. (Default 0 Hz).
high_freq (float) – highest band edge of mel filters. (Default samplerate/2).
scale (str) – monotonicity behavior of the filter banks. (Default is “constant”).
fb_type (str) – the filter banks type. (Default is “mel”)
conversion_approach (str) – mel scale conversion approach. (Default is “Oshaghnessy”).

Returns

(numpy.ndarray) : array of size nfilts * (nfft/2 + 1) containing filter bank. Each row holds 1 filter.
(numpy.ndarray) : array of center frequencies

Return type

(tuple)

Tip

scale : can take the following options [“constant”, “ascendant”, “descendant”].
fb_type : can take the following options [“mel”, “lin”].
conversion_approach : can take the following options [“Oshaghnessy”, “Beranek”, “Lindsay”]. Note that the use of different options than the default can lead to unexpected behavior/issues.

spafe.fbanks.mel_fbanks.mel_filter_banks(nfilts: int = 24, nfft: int = 512, fs: int = 16000, low_freq: float = 0, high_freq: Optional[float] = None, scale: typing_extensions.Literal[ascendant, descendant, constant] = 'constant', conversion_approach: typing_extensions.Literal[Oshaghnessy, Lindsay] = 'Oshaghnessy')[source]#

Compute Mel-filter banks.The filters are stored in the rows, the columns correspond to fft bins.

Parameters

nfilts (int) – the number of filters in the filter bank. (Default 20).
nfft (int) – the FFT size. (Default is 512).
fs (int) – sample rate/ sampling frequency of the signal. (Default 16000 Hz).
low_freq (float) – lowest band edge of mel filters. (Default 0 Hz).
high_freq (float) – highest band edge of mel filters. (Default samplerate/2).
scale (str) – monotonicity behavior of the filter banks. (Default is “constant”).
conversion_approach (str) – mel scale conversion approach. (Default is “Oshaghnessy”).

Returns

(numpy.ndarray) : array of size nfilts * (nfft/2 + 1) containing filter bank. Each row holds 1 filter.
(numpy.ndarray) : array of center frequencies

Return type

(tuple)

Tip

scale : can take the following options [“constant”, “ascendant”, “descendant”].
conversion_approach : can take the following options [“Oshaghnessy”, “Beranek”, “Lindsay”]. Note that the use of different options than the default can lead to unexpected behavior/issues.

Examples

import numpy as np
from spafe.utils.converters import mel2hz
from spafe.utils.vis import show_fbanks
from spafe.fbanks.mel_fbanks import mel_filter_banks

# init var
fs = 8000
nfilt = 7
nfft = 1024
low_freq = 0
high_freq = fs / 2

# compute freqs for xaxis
mhz_freqs = np.linspace(low_freq, high_freq, nfft //2+1)

for scale, label in [("constant", ""), ("ascendant", "Ascendant "), ("descendant", "Descendant ")]:
    # compute fbank
    mel_fbanks_mat, mel_freqs = mel_filter_banks(nfilts=nfilt,
                                                 nfft=nfft,
                                                 fs=fs,
                                                 low_freq=low_freq,
                                                 high_freq=high_freq,
                                                scale=scale)

    # visualize fbank
    show_fbanks(
        mel_fbanks_mat,
        [mel2hz(freq) for freq in mel_freqs],
        mhz_freqs,
        label + "Mel Filter Bank",
        ylabel="Weight",
        x1label="Frequency / Hz",
        x2label="Frequency / Mel",
        figsize=(14, 5),
        fb_type="mel")

spafe.fbanks.mel_fbanks.inverse_mel_filter_banks(nfilts: int = 24, nfft: int = 512, fs: int = 16000, low_freq: float = 0, high_freq: Optional[float] = None, scale: typing_extensions.Literal[ascendant, descendant, constant] = 'constant', conversion_approach: typing_extensions.Literal[Oshaghnessy, Lindsay] = 'Oshaghnessy')[source]#

Compute inverse Mel-filter banks. The filters are stored in the rows, the columns correspond to fft bins.

Parameters

nfilt (int) – the number of filters in the filter bank. (Default 20).
nfft (int) – the FFT size. (Default is 512).
fs (int) – sample rate/ sampling frequency of the signal. (Default 16000 Hz).
low_freq (float) – lowest band edge of mel filters. (Default 0 Hz).
high_freq (float) – highest band edge of mel filters. (Default samplerate/2).
scale (str) – monotonicity behavior of the filter banks. (Default is “constant”).
conversion_approach (str) – mel scale conversion approach. (Default is “Oshaghnessy”).

Returns

(numpy.ndarray) : array of size nfilts * (nfft/2 + 1) containing filter bank. Each row holds 1 filter.
(numpy.ndarray) : array of center frequencies

Return type

(tuple)

Tip

scale : can take the following options [“constant”, “ascendant”, “descendant”].
conversion_approach : can take the following options [“Oshaghnessy”, “Beranek”, “Lindsay”]. Note that the use of different options than the default can lead to unexpected behavior/issues.

Examples

import numpy as np
from spafe.utils.converters import mel2hz
from spafe.utils.vis import show_fbanks
from spafe.fbanks.mel_fbanks import inverse_mel_filter_banks

# init var
fs = 8000
nfilt = 7
nfft = 1024
low_freq = 0
high_freq = fs / 2

# compute freqs for xaxis
mhz_freqs = np.linspace(low_freq, high_freq, nfft //2+1)

for scale, label in [("constant", ""), ("ascendant", "Ascendant "), ("descendant", "Descendant ")]:
    # compute fbank
    imel_fbanks_mat, imel_freqs = inverse_mel_filter_banks(nfilts=nfilt,
                                                           nfft=nfft,
                                                           fs=fs,
                                                           low_freq=low_freq,
                                                           high_freq=high_freq,
                                                           scale=scale)

    # visualize fbank
    show_fbanks(
        imel_fbanks_mat,
        [mel2hz(freq) for freq in imel_freqs],
        mhz_freqs,
        label + "Inverse Mel Filter Bank",
        ylabel="Weight",
        x1label="Frequency / Hz",
        x2label="Frequency / Mel",
        figsize=(14, 5),
        fb_type="mel")