Pitch

class audioflux.PitchYIN(samplate=32000, low_fre=27.0, high_fre=2000.0, radix2_exp=12, slide_length=1024, auto_length=2048)

Pitch is estimated using time-domain differential autocorrelation. 1

1: Alain de Cheveigne, Hideki Kawahara. “YIN, a fundamental frequency estimator for speech and music.” 2002 Acoustical Societyof America.

Parameters

samplate: int: Sampling rate of the incoming audio.
low_fre: float: Lowest frequency. Default is 27.0.
high_fre: float: Highest frequency. Default is 2000.0.
radix2_exp: int: fft_length=2**radix2_exp
slide_length: int: Window sliding length.
auto_length: int: Auto correlation length. Default is 2048.

Examples

Read voice audio data

>>> import audioflux as af
>>> audio_path = af.utils.sample_path('voice')
>>> audio_arr, sr = af.read(audio_path)

Extract pitch

>>> pitch_obj = af.PitchYIN(samplate=sr)
>>> fre_arr, v1_arr, v2_arr = pitch_obj.pitch(audio_arr)

Show pitch plot

>>> import matplotlib.pyplot as plt
>>> times = np.arange(fre_arr.shape[-1]) * (pitch_obj.slide_length / sr)
>>> fig, ax = plt.subplots(nrows=2, sharex=True)
>>> ax[0].set_title('PitchYIN')
>>> af.display.fill_wave(audio_arr, samplate=sr, axes=ax[0])
>>> ax[1].scatter(times, fre_arr, s=2)

Methods

`cal_time_length`(data_length)	Calculate the length of a frame from audio data.
`pitch`(data_arr)	Compute pitch
`set_thresh`(thresh)	Set thresh

set_thresh(thresh)

Set thresh

Parameters

thresh: float

cal_time_length(data_length)

Calculate the length of a frame from audio data.

fft_length = 2 ** radix2_exp
(data_length - fft_length) // slide_length + 1

Parameters

data_length: int: The length of the data to be calculated.

Returns

out: int

pitch(data_arr)

Compute pitch

Parameters

data_arr: np.ndarray [shape=(…, n)]: Input audio array

Returns

fre_arr: np.ndarray [shape=(…, time)]
value1_arr: np.ndarray [shape=(…, time)]
value2_arr: np.ndarray [shape=(…, time)]

class audioflux.PitchPEF(samplate=32000, low_fre=32.0, high_fre=2000.0, cut_fre=4000.0, radix2_exp=12, slide_length=1024, window_type=WindowType.HAMM, alpha=10.0, beta=0.5, gamma=1.8)

Pitch Estimation Filter(PEF). A pitch estimation filter is designed, and pitch is estimated by performing cross-correlation operations in the frequency domain. 2

2: Gonzalez, Sira, and Mike Brookes. “A Pitch Estimation Filter robust to high levels of noise (PEFAC).” 19th European Signal Processing Conference. Barcelona, 2011, pp. 451–455.

Parameters

samplate: int

Sampling rate of the incoming audio.

low_fre: float

Lowest frequency. Default is 32.0.

high_fre: float

Highest frequency. Default is 2000.0.

cut_fre: float

Cut frequency. Default is 4000.0, and must be greater than high_fre.

radix2_exp: int

fft_length=2**radix2_exp

slide_length: int

Window sliding length.

window_type: WindowType

Window type for each frame.

See: type.WindowType

alpha: float, > 0

alpha. Default if 10.0..

beta: float, 0~1

beta. Default if 0.5..

gamma: float, > 1

gamma. Default if 1.8.

Examples

Read voice audio data

>>> import audioflux as af
>>> audio_path = af.utils.sample_path('voice')
>>> audio_arr, sr = af.read(audio_path)

Extract pitch

>>> pitch_obj = af.PitchPEF(samplate=sr)
>>> fre_arr = pitch_obj.pitch(audio_arr)

Show pitch plot

>>> import matplotlib.pyplot as plt
>>> times = np.arange(fre_arr.shape[-1]) * (pitch_obj.slide_length / sr)
>>> fig, ax = plt.subplots(nrows=2, sharex=True)
>>> ax[0].set_title('PitchPEF')
>>> af.display.fill_wave(audio_arr, samplate=sr, axes=ax[0])
>>> ax[1].scatter(times, fre_arr, s=2)

Methods

`cal_time_length`(data_length)	Calculate the length of a frame from audio data.
`pitch`(data_arr)	Compute pitch
`set_filter_params`(alpha, beta, gamma)	Set filter params

cal_time_length(data_length)

Calculate the length of a frame from audio data.

fft_length = 2 ** radix2_exp
(data_length - fft_length) // slide_length + 1

Parameters

data_length: int: The length of the data to be calculated.

Returns

out: int

set_filter_params(alpha, beta, gamma)

Set filter params

Parameters

alpha: float: alpha
beta: float: beta
gamma: float: gamma

pitch(data_arr)

Compute pitch

Parameters

data_arr: np.ndarray [shape=(…, n)]: Input audio array

Returns

fre_arr: np.ndarray [shape=(…, time)]

class audioflux.PitchCEP(samplate=32000, low_fre=32.0, high_fre=2000.0, radix2_exp=12, slide_length=1024, window_type=WindowType.HAMM)

Cepstrum Pitch Determination(CEP). Pitch is estimated by performing a second FFT transformation on the spectrum and using cepstral analysis. 3

3: Noll, Michael A. “Cepstrum Pitch Determination.” The Journal of the Acoustical Society of America. Vol. 31, No. 2, 1967, pp. 293–309.

Parameters

samplate: int

Sampling rate of the incoming audio.

low_fre: float

Lowest frequency. Default is 32.0.

high_fre: float

Highest frequency. Default is 2000.0.

radix2_exp: int

fft_length=2**radix2_exp

slide_length: int

Window sliding length.

window_type: WindowType

Window type for each frame.

See: type.WindowType

Examples

Read voice audio data

>>> import audioflux as af
>>> audio_path = af.utils.sample_path('voice')
>>> audio_arr, sr = af.read(audio_path)

Extract pitch

>>> pitch_obj = af.PitchCEP(samplate=sr)
>>> fre_arr = pitch_obj.pitch(audio_arr)

Show pitch plot

>>> import matplotlib.pyplot as plt
>>> times = np.arange(fre_arr.shape[-1]) * (pitch_obj.slide_length / sr)
>>> fig, ax = plt.subplots(nrows=2, sharex=True)
>>> ax[0].set_title('PitchCEP')
>>> af.display.fill_wave(audio_arr, samplate=sr, axes=ax[0])
>>> ax[1].scatter(times, fre_arr, s=2)

Methods

`cal_time_length`(data_length)	Calculate the length of a frame from audio data.
`pitch`(data_arr)	Compute pitch

cal_time_length(data_length)

Calculate the length of a frame from audio data.

fft_length = 2 ** radix2_exp
(data_length - fft_length) // slide_length + 1

Parameters

data_length: int: The length of the data to be calculated.

Returns

out: int

pitch(data_arr)

Compute pitch

Parameters

data_arr: np.ndarray [shape=(…, n)]: Input audio array

Returns

fre_arr: np.ndarray [shape=(…, time)]

class audioflux.PitchHPS(samplate=32000, low_fre=32.0, high_fre=2000.0, radix2_exp=12, slide_length=1024, window_type=WindowType.HAMM, harmonic_count=5)

Dot-Harmonic Spectrum(HPS). Pitch is estimated by adopting dot operations on the harmonics of the spectrum. 4

4: Tamara Smyth. “Music270a: Signal Analysis.” 2019, Department of Music, University of California.

Parameters

samplate: int

Sampling rate of the incoming audio.

low_fre: float

Lowest frequency. Default is 32.0.

high_fre: float

Highest frequency. Default is 2000.0.

radix2_exp: int

fft_length=2**radix2_exp

slide_length: int

Window sliding length.

window_type: WindowType

Window type for each frame.

See: type.WindowType

harmonic_count: int

Harmonic count. Default is 5.

Examples

Read voice audio data

>>> import audioflux as af
>>> audio_path = af.utils.sample_path('voice')
>>> audio_arr, sr = af.read(audio_path)

Extract pitch

>>> pitch_obj = af.PitchHPS(samplate=sr)
>>> fre_arr = pitch_obj.pitch(audio_arr)

Show pitch plot

>>> import matplotlib.pyplot as plt
>>> times = np.arange(fre_arr.shape[-1]) * (pitch_obj.slide_length / sr)
>>> fig, ax = plt.subplots(nrows=2, sharex=True)
>>> ax[0].set_title('PitchHPS')
>>> af.display.fill_wave(audio_arr, samplate=sr, axes=ax[0])
>>> ax[1].scatter(times, fre_arr, s=2)

Methods

`cal_time_length`(data_length)	Calculate the length of a frame from audio data.
`pitch`(data_arr)	Compute pitch

cal_time_length(data_length)

Calculate the length of a frame from audio data.

fft_length = 2 ** radix2_exp
(data_length - fft_length) // slide_length + 1

Parameters

data_length: int: The length of the data to be calculated.

Returns

out: int

pitch(data_arr)

Compute pitch

Parameters

data_arr: np.ndarray [shape=(…, n)]: Input audio array

Returns

fre_arr: np.ndarray [shape=(…, time)]

class audioflux.PitchLHS(samplate=32000, low_fre=32.0, high_fre=2000.0, radix2_exp=12, slide_length=1024, window_type=WindowType.HAMM, harmonic_count=5)

Log-Harmonic Summation(LHS). Pitch is estimated by adopting sum operations on the harmonics of the spectrum. 5

5: Hermes, Dik J. “Measurement of Pitch by Subharmonic Summation.”The Journal of the Acoustical Society of America. Vol. 83, No. 1, 1988, pp. 257–264.

Parameters

samplate: int

Sampling rate of the incoming audio.

low_fre: float

Lowest frequency. Default is 32.0.

high_fre: float

Highest frequency. Default is 2000.0.

radix2_exp: int

fft_length=2**radix2_exp

slide_length: int

Window sliding length.

window_type: WindowType

Window type for each frame.

See: type.WindowType

harmonic_count: int

Harmonic count. Default is 5.

Examples

Read voice audio data

>>> import audioflux as af
>>> audio_path = af.utils.sample_path('voice')
>>> audio_arr, sr = af.read(audio_path)

Extract pitch

>>> pitch_obj = af.PitchLHS(samplate=sr)
>>> fre_arr = pitch_obj.pitch(audio_arr)

Show pitch plot

>>> import matplotlib.pyplot as plt
>>> times = np.arange(fre_arr.shape[-1]) * (pitch_obj.slide_length / sr)
>>> fig, ax = plt.subplots(nrows=2, sharex=True)
>>> ax[0].set_title('PitchLHS')
>>> af.display.fill_wave(audio_arr, samplate=sr, axes=ax[0])
>>> ax[1].scatter(times, fre_arr, s=2)

Methods

`cal_time_length`(data_length)	Calculate the length of a frame from audio data.
`pitch`(data_arr)	Compute pitch

cal_time_length(data_length)

Calculate the length of a frame from audio data.

fft_length = 2 ** radix2_exp
(data_length - fft_length) / slide_length + 1

Parameters

data_length: int: The length of the data to be calculated.

Returns

out: int

pitch(data_arr)

Compute pitch

Parameters

data_arr: np.ndarray [shape=(…, n)]: Input audio array

Returns

fre_arr: np.ndarray [shape=(…, time)]

class audioflux.PitchNCF(samplate=32000, low_fre=32.0, high_fre=2000.0, radix2_exp=12, slide_length=1024, window_type=WindowType.RECT)

Normalized Correlation Function(NCF). Pitch is estimated using normalized time-domain autocorrelation. 6

6: Atal, B.S. “Automatic Speaker Recognition Based on Pitch Contours.” The Journal of the Acoustical Society of America. Vol. 52, No. 6B, 1972, pp. 1687–1697.`

Parameters

samplate: int

Sampling rate of the incoming audio.

low_fre: float

Lowest frequency. Default is 32.0.

high_fre: float

Highest frequency. Default is 2000.0.

radix2_exp: int

fft_length=2**radix2_exp

slide_length: int

Window sliding length.

window_type: WindowType

Window type for each frame.

See: type.WindowType

Examples

Read voice audio data

>>> import audioflux as af
>>> audio_path = af.utils.sample_path('voice')
>>> audio_arr, sr = af.read(audio_path)

Extract pitch

>>> pitch_obj = af.PitchNCF(samplate=sr)
>>> fre_arr = pitch_obj.pitch(audio_arr)

Show pitch plot

>>> import matplotlib.pyplot as plt
>>> times = np.arange(fre_arr.shape[-1]) * (pitch_obj.slide_length / sr)
>>> fig, ax = plt.subplots(nrows=2, sharex=True)
>>> ax[0].set_title('PitchNCF')
>>> af.display.fill_wave(audio_arr, samplate=sr, axes=ax[0])
>>> ax[1].scatter(times, fre_arr, s=2)

Methods

`cal_time_length`(data_length)	Calculate the length of a frame from audio data.
`pitch`(data_arr)	Compute pitch

cal_time_length(data_length)

Calculate the length of a frame from audio data.

fft_length = 2 ** radix2_exp
(data_length - fft_length) / slide_length + 1

Parameters

data_length: int: The length of the data to be calculated.

Returns

out: int

pitch(data_arr)

Compute pitch

Parameters

data_arr: np.ndarray [shape=(…, n)]: Input audio array

Returns

fre_arr: np.ndarray [shape=(…, time)]

class audioflux.PitchSTFT(samplate=32000, low_fre=32.0, high_fre=2000.0, radix2_exp=12, slide_length=1024, window_type=WindowType.HAMM)

Pitch STFT algorithm

Parameters

samplate: int

Sampling rate of the incoming audio.

low_fre: float

Lowest frequency. Default is 32.0.

high_fre: float

Highest frequency. Default is 2000.0.

radix2_exp: int

fft_length=2**radix2_exp

slide_length: int

Window sliding length.

window_type: WindowType

Window type for each frame.

See: type.WindowType

Examples

Read voice audio data

>>> import audioflux as af
>>> audio_path = af.utils.sample_path('voice')
>>> audio_arr, sr = af.read(audio_path)

Extract pitch

>>> pitch_obj = af.PitchSTFT(samplate=sr)
>>> fre_arr, db_arr = pitch_obj.pitch(audio_arr)

Show pitch plot

>>> import matplotlib.pyplot as plt
>>> times = np.arange(fre_arr.shape[-1]) * (pitch_obj.slide_length / sr)
>>> fig, ax = plt.subplots(nrows=2, sharex=True)
>>> ax[0].set_title('PitchSTFT')
>>> af.display.fill_wave(audio_arr, samplate=sr, axes=ax[0])
>>> ax[1].scatter(times, fre_arr, s=2)

Methods

`cal_time_length`(data_length)	Calculate the length of a frame from audio data.
`pitch`(data_arr)	Compute pitch

cal_time_length(data_length)

Calculate the length of a frame from audio data.

fft_length = 2 ** radix2_exp
(data_length - fft_length) // slide_length + 1

Parameters

data_length: int: The length of the data to be calculated.

Returns

out: int

pitch(data_arr)

Compute pitch

Parameters

data_arr: np.ndarray [shape=(…, n)]: Input audio array

Returns

fre_arr: np.ndarray [shape=(…, time)]
db_arr: np.ndarray [shape=(…, time)]

class audioflux.PitchFFP(samplate=32000, low_fre=32.0, high_fre=2000.0, radix2_exp=12, slide_length=1024, window_type=WindowType.HAMM)

Pitch FFP algorithm

Parameters

samplate: int

Sampling rate of the incoming audio.

low_fre: float

Lowest frequency. Default is 32.0.

high_fre: float

Highest frequency. Default is 2000.0.

radix2_exp: int

fft_length=2**radix2_exp

slide_length: int

Window sliding length.

window_type: WindowType

Window type for each frame.

See: type.WindowType

Examples

Read voice audio data

>>> import audioflux as af
>>> audio_path = af.utils.sample_path('voice')
>>> audio_arr, sr = af.read(audio_path)

Extract pitch

>>> pitch_obj = af.PitchFFP(samplate=sr, radix2_exp=12, low_fre=27, high_fre=2000,
>>>                         slide_length=1024, window_type=af.type.WindowType.HAMM)
>>> fre_arr, db_arr, extra_data_dic = pitch_obj.pitch(
>>>     audio_arr,
>>>     has_corr_data=True,
>>>     has_cut_data=True,
>>>     has_flag_data=True,
>>>     has_light_data=True,
>>>     has_temporal_data=True)
>>> cut_corr_arr, cut_db_arr, cut_height_arr, cut_length_arr = extra_data_dic['cut_data']
>>> flag_arr, = extra_data_dic['flag_data']
>>> light_arr, = extra_data_dic['light_data']
>>> avg_temp_arr, max_temp_arr, percent_temp_arr = extra_data_dic['temporal_data']

Show pitch plot

>>> import matplotlib.pyplot as plt
>>> times = np.arange(fre_arr.shape[-1]) * (pitch_obj.slide_length / sr)
>>>
>>> # Show pitch plot
>>> fig, ax = plt.subplots(nrows=3, sharex=True)
>>> ax[0].set_title('PitchFFP')
>>> af.display.fill_wave(audio_arr, axes=ax[0])
>>> ax[1].scatter(times, fre_arr, s=2)
>>> ax[2].scatter(times, db_arr, s=2)
>>>
>>> # Show cut data plot
>>> fig, ax = plt.subplots(nrows=4, sharex=True)
>>> ax[0].set_title('FFP Corr Data')
>>> af.display.fill_wave(audio_arr, axes=ax[0])
>>> for i in range(cut_corr_arr.shape[0]):
>>>     ax[1].scatter(times, cut_corr_arr[i], s=2)
>>> for i in range(cut_db_arr.shape[0]):
>>>     ax[2].scatter(times, cut_db_arr[i], s=2)
>>> for i in range(cut_height_arr.shape[0]):
>>>     ax[3].scatter(times, cut_height_arr[i], s=2)
>>>
>>> # Show flag and light data plot
>>> fig, ax = plt.subplots(nrows=3, sharex=True)
>>> ax[0].set_title('FFP Flag and Light Data')
>>> af.display.fill_wave(audio_arr, axes=ax[0])
>>> ax[1].scatter(times, flag_arr, s=2, label='Flag')
>>> ax[1].legend()
>>> ax[2].scatter(times, light_arr, s=2, label='Light')
>>> ax[2].legend()
>>>
>>> # Show temporal data plot
>>> fig, ax = plt.subplots(nrows=4, sharex=True)
>>> ax[0].set_title('FFP Temporal Data')
>>> af.display.fill_wave(audio_arr, axes=ax[0])
>>> ax[1].scatter(times, avg_temp_arr, s=2)
>>> ax[2].scatter(times, max_temp_arr, s=2)
>>> ax[3].scatter(times, percent_temp_arr, s=2)

Methods

`cal_time_length`(data_length)	Calculate the length of a frame from audio data.
`pitch`(data_arr[, has_corr_data, ...])	Compute pitch
`set_temp_base`(temp_base)	Set temproal base

cal_time_length(data_length)

Calculate the length of a frame from audio data.

fft_length = 2 ** radix2_exp
(data_length - fft_length) // slide_length + 1

Parameters

data_length: int: The length of the data to be calculated.

Returns

out: int

set_temp_base(temp_base)

Set temproal base

Parameters

temp_base: float: temproal base

pitch(data_arr, has_corr_data=False, has_cut_data=False, has_flag_data=False, has_light_data=False, has_temporal_data=False)

Compute pitch

Parameters

data_arr: np.ndarray [shape=(…, n)]

Input audio array

has_corr_data: bool

If true, the extra_data_dic in the result will contain corr_data information.

corr_data includes four arrays: corr_arr, db_arr, height_arr, and len_arr.

has_cut_data: bool

If true, the extra_data_dic in the result will contain cut_data information. cut_data contains only the first four sets of data from corr_data.

cut_data includes four arrays: corr_arr, db_arr, height_arr, and len_arr.

has_flag_data: bool

If true, the extra_data_dic in the result will contain flag_data information.

flag_data contains one array, flag_arr.

has_light_data: bool

If true, the extra_data_dic in the result will contain light_data information.

light_data contains one array, light_arr.

has_temporal_data: bool

If true, the extra_data_dic in the result will contain temporal_data information.

temporal_data includes three arrays: avg_temp_arr, max_temp_arr, percent_temp_arr.

Returns

fre_arr: np.ndarray [shape=(…, time)]

frequency array. No data.

db_arr: np.ndarray [shape=(…, time)]

db array

extra_data_dic: (optional) dict

If all has_xx_data values are False, the extra_data_dic will not be included in the returned result.

If any has_xx_data is True, the extra_data_dic will include the corresponding data.