U %dʏ@s@ddlZddlmZddlZddlZddlmZdddddd d d d d g Zeeej j Z dZ dZ dZdZdZdZe eeeegZddZeedddZeeeeedddZeee ejeedddZeee edd d!Zeee e e ee eeeeefd"d#d$Zeeeeee eee e ee eeefd% d&d'Zeeed(d)d*Zd+d,d-d.d/d0d-d1d2d2d2d3d2d4efee ee e e e e e eeee eeeed5d6d Ze e d7d8dZ eed7d9dZ!e e d:d;dZ"eed:dd Z$e e e e eed?d@d Z%eee e e e e e eeefdA dBdZ&d+d,d-d.d/d0d-d4dCd-dDd1d2d2d2d3d2d4d4d2d2dEdFd.efee ee e e e e ee e ee eeee eeeeee e e eedGdHd Z'eeedIdJdKZ(ee edLdMdNZ)d+dOd,d-d.d/d0d-d4dCdPd-dDd1d2d2d2d3d2d4d4dEdFd.efee e ee e e e e ee ee ee eeee eeee e e eedQdRd Z*dS)SN)Tuple)Tensor get_mel_banksinverse_mel_scaleinverse_mel_scale_scalar mel_scalemel_scale_scalar spectrogramfbankmfccvtln_warp_freqvtln_warp_mel_freqgMbP?ZhammingZhanningZpoveyZ rectangularZblackmancCstj||dS)Ndevicedtype)EPSILONtorr?/tmp/pip-unpacked-wheel-lbdmvq91/torchaudio/compliance/kaldi.py _get_epsilon#sr)xreturncCs|dkr dSd|dS)z6Returns the smallest power of 2 that is greater than xr) bit_length)rrrr_next_power_of_2'sr)waveform window_size window_shift snip_edgesrc Cs|dkst|d}||d|df}|rd||krRtjd|j|jdSd|||}nzt|dg}||d|}|d|d}|} |dkr|| d} tj | || fdd}ntj || d| fdd}||f} | | |S)aGiven a waveform (1D tensor of size ``num_samples``), it returns a 2D tensor (m, ``window_size``) representing how the window is shifted along the waveform. Each row is a frame. Args: waveform (Tensor): Tensor of size ``num_samples`` window_size (int): Frame length window_shift (int): Frame shift snip_edges (bool): If True, end effects will be handled by outputting only frames that completely fit in the file, and the number of frames depends on the frame_length. If False, the number of frames depends only on the frame_shift, and we reflect the data at the ends. Returns: Tensor: 2D tensor of size (m, ``window_size``) where each row is a frame rr)rrrrrNdim) r"AssertionErrorsizeZstridetorchemptyrrZflipcatZ as_strided) rrrrZ num_samplesstridesmZreversed_waveformpadZ pad_rightZpad_leftZsizesrrr _get_strided,s" r+) window_typerblackman_coeffrrrcCs|tkrtj|d||dS|tkr8tj|ddd||dS|tkrXtj|d||ddS|tkrptj|||dS|t krdt j |d }tj |||d}|d t ||d |t d||j||dStd |d S) z6Returns a window function with the given type and sizeF)periodicrrgHzG?gq= ףp?)r.alphabetarrg333333?rrr?zInvalid window type N)HANNINGr%Z hann_windowHAMMINGZhamming_windowPOVEYpow RECTANGULARZonesBLACKMANmathpiarangecosr Exception)r,rr-rrawindow_functionrrr_feature_window_functionVs*r?) strided_inputepsilon energy_floorrcCsT|j|j}}t|dd|}|dkr6|St|tjt|||dS)zZ padding_rightrrr _get_windowsD &    rf)rG subtract_meanrcCs$|r tj|ddd}||}|S)Nrr!)r%rarb)rGrgZ col_meansrrr_subtract_column_meansrhgzG?r[rCrOg9@g$@g ףp= ?Tg@@F)rr-rIrWrBrLrK min_durationrNrVrXrMrJrrgr,rc Cs|j|j}}t||}t||| ||| |\}}}}t||| krNtdSt||||||| | ||| | \}}tj |}t | d| }||dddf<t||}|S)a Create a spectrogram from a raw audio signal. This matches the input/output of Kaldi's compute-spectrogram-feats. Args: waveform (Tensor): Tensor of audio of size (c, n) where c is in the range [0,2) blackman_coeff (float, optional): Constant coefficient for generalized Blackman window. (Default: ``0.42``) channel (int, optional): Channel to extract (-1 -> expect mono, 0 -> left, 1 -> right) (Default: ``-1``) dither (float, optional): Dithering constant (0.0 means no dither). If you turn this off, you should set the energy_floor option, e.g. to 1.0 or 0.1 (Default: ``0.0``) energy_floor (float, optional): Floor on energy (absolute, not relative) in Spectrogram computation. Caution: this floor is applied to the zeroth component, representing the total signal energy. The floor on the individual spectrogram elements is fixed at std::numeric_limits::epsilon(). (Default: ``1.0``) frame_length (float, optional): Frame length in milliseconds (Default: ``25.0``) frame_shift (float, optional): Frame shift in milliseconds (Default: ``10.0``) min_duration (float, optional): Minimum duration of segments to process (in seconds). (Default: ``0.0``) preemphasis_coefficient (float, optional): Coefficient for use in signal preemphasis (Default: ``0.97``) raw_energy (bool, optional): If True, compute energy before preemphasis and windowing (Default: ``True``) remove_dc_offset (bool, optional): Subtract mean from waveform on each frame (Default: ``True``) round_to_power_of_two (bool, optional): If True, round window size to power of two by zero-padding input to FFT. (Default: ``True``) sample_frequency (float, optional): Waveform data sample frequency (must match the waveform file, if specified there) (Default: ``16000.0``) snip_edges (bool, optional): If True, end effects will be handled by outputting only frames that completely fit in the file, and the number of frames depends on the frame_length. If False, the number of frames depends only on the frame_shift, and we reflect the data at the ends. (Default: ``True``) subtract_mean (bool, optional): Subtract mean of each feature file [CMS]; not recommended to do it this way. (Default: ``False``) window_type (str, optional): Type of window ('hamming'|'hanning'|'povey'|'rectangular'|'blackman') (Default: ``'povey'``) Returns: Tensor: A spectrogram identical to what Kaldi would output. The shape is (m, ``padded_window_size // 2 + 1``) where m is calculated in _get_strided r@N)rrrrUrSr%r&rffftrfftrDabsr5rFrh)rr-rIrWrBrLrKrirNrVrXrMrJrrgr,rrrArrrTr@rerkZpower_spectrumrrrr s@4     )mel_freqrcCsdt|ddSN@@rO)r8exprnrrrr@scCsd|ddSro)rrrsrrrrDs)freqrcCsdtd|dSNrqrOrp)r8rFrtrrrrHscCsdd|dSru)rFrvrrrrLs)vtln_low_cutoffvtln_high_cutofflow_freq high_freqvtln_warp_factorrtrcCs||kstd||ks td|td|}|td|}d|}||} ||} ||krd||ksht| |||} || ||} t|} t||t||B}t||}t||}t||}|| |||| |<|||| |<|| |||| |<||| |<| S)a*This computes a VTLN warping function that is not the same as HTK's one, but has similar inputs (this function has the advantage of never producing empty bins). This function computes a warp function F(freq), defined between low_freq and high_freq inclusive, with the following properties: F(low_freq) == low_freq F(high_freq) == high_freq The function is continuous and piecewise linear with two inflection points. The lower inflection point (measured in terms of the unwarped frequency) is at frequency l, determined as described below. The higher inflection point is at a frequency h, determined as described below. If l <= f <= h, then F(f) = f/vtln_warp_factor. If the higher inflection point (measured in terms of the unwarped frequency) is at h, then max(h, F(h)) == vtln_high_cutoff. Since (by the last point) F(h) == h/vtln_warp_factor, then max(h, h/vtln_warp_factor) == vtln_high_cutoff, so h = vtln_high_cutoff / max(1, 1/vtln_warp_factor). = vtln_high_cutoff * min(1, vtln_warp_factor). If the lower inflection point (measured in terms of the unwarped frequency) is at l, then min(l, F(l)) == vtln_low_cutoff This implies that l = vtln_low_cutoff / min(1, 1/vtln_warp_factor) = vtln_low_cutoff * max(1, vtln_warp_factor) Args: vtln_low_cutoff (float): Lower frequency cutoffs for VTLN vtln_high_cutoff (float): Upper frequency cutoffs for VTLN low_freq (float): Lower frequency cutoffs in mel computation high_freq (float): Upper frequency cutoffs in mel computation vtln_warp_factor (float): Vtln warp factor freq (Tensor): given frequency in Hz Returns: Tensor: Freq after vtln warp z7be sure to set the vtln_low option higher than low_freqzFbe sure to set the vtln_high option lower than high_freq [or negative]rO)r#rDminr%Z empty_likeltgtge)rwrxryrzr{rtlhZscaleZFlZFhZ scale_leftZ scale_rightresZoutside_low_high_freqZbefore_lZbefore_hZafter_hrrrr Ps(,     )rwrxrzr{rnrc Cstt|||||t|S)a Args: vtln_low_cutoff (float): Lower frequency cutoffs for VTLN vtln_high_cutoff (float): Upper frequency cutoffs for VTLN low_freq (float): Lower frequency cutoffs in mel computation high_freq (float): Upper frequency cutoffs in mel computation vtln_warp_factor (float): Vtln warp factor mel_freq (Tensor): Given frequency in Mel Returns: Tensor: ``mel_freq`` after vtln warp )rr r)rwrxryrzr{rnrrrr s) num_binswindow_length_padded sample_freqryrzvtln_low vtln_highr{rcCsZ|dkstd|ddks t|d}d|} |dkr@|| 7}d|krT| krxnn d|krl| krxnn||kstd||| ||} t|} t|} | | |d} |dkr|| 7}|d ks||kr|kr nn$d|kr|kr nn ||kstd ||||t|d}| || }| |d | }| |d | }|d krt||||||}t||||||}t||||||}t|}t| t|d}||||}||||}|d krt t dt ||}nRt |}t ||t||@}t ||t||@}||||<||||<||fS) z Returns: (Tensor, Tensor): The tuple consists of ``bins`` (which is melbank of size (``num_bins``, ``num_fft_bins``)) and ``center_freqs`` (which is center frequencies of bins of size (``num_bins``)). zMust have at least 3 mel binsrrr1rCzBBad values in options: low-freq {} and high-freq {} vs. nyquist {}rrOzXBad values in options: vtln-low {} and vtln-high {}, versus low-freq {} and high-freq {}rj)r#rPrr%r:rbr rrrDzerosr|Z zeros_liker~ler})rrrryrzrrr{Z num_fft_binsZnyquistZ fft_bin_widthZ mel_low_freqZ mel_high_freqZmel_freq_deltabinZleft_melZ center_melZ right_melZ center_freqsZmelZup_slopeZ down_slopeZbinsZup_idxZdown_idxrrrrs            g4@g@gY@)rr-rIrWrBrLrKrz htk_compatryri num_mel_binsrNrVrXrMrJrrg use_energy use_log_fbank use_powerrr vtln_warpr,rc$ Cs:|j|j}}t||||||| \}}}}t|| |krJtjd||dSt|||||||| |||| \}} tj| }!|r|! d}!t | ||| ||||\}"}#|"j ||d}"tj jj|"dddd}"t|!|"j}"|rt|"t||}"|r,| d} |rtj|"| fdd}"ntj| |"fdd}"t|"|}"|"S) abCreate a fbank from a raw audio signal. This matches the input/output of Kaldi's compute-fbank-feats. Args: waveform (Tensor): Tensor of audio of size (c, n) where c is in the range [0,2) blackman_coeff (float, optional): Constant coefficient for generalized Blackman window. (Default: ``0.42``) channel (int, optional): Channel to extract (-1 -> expect mono, 0 -> left, 1 -> right) (Default: ``-1``) dither (float, optional): Dithering constant (0.0 means no dither). If you turn this off, you should set the energy_floor option, e.g. to 1.0 or 0.1 (Default: ``0.0``) energy_floor (float, optional): Floor on energy (absolute, not relative) in Spectrogram computation. Caution: this floor is applied to the zeroth component, representing the total signal energy. The floor on the individual spectrogram elements is fixed at std::numeric_limits::epsilon(). (Default: ``1.0``) frame_length (float, optional): Frame length in milliseconds (Default: ``25.0``) frame_shift (float, optional): Frame shift in milliseconds (Default: ``10.0``) high_freq (float, optional): High cutoff frequency for mel bins (if <= 0, offset from Nyquist) (Default: ``0.0``) htk_compat (bool, optional): If true, put energy last. Warning: not sufficient to get HTK compatible features (need to change other parameters). (Default: ``False``) low_freq (float, optional): Low cutoff frequency for mel bins (Default: ``20.0``) min_duration (float, optional): Minimum duration of segments to process (in seconds). (Default: ``0.0``) num_mel_bins (int, optional): Number of triangular mel-frequency bins (Default: ``23``) preemphasis_coefficient (float, optional): Coefficient for use in signal preemphasis (Default: ``0.97``) raw_energy (bool, optional): If True, compute energy before preemphasis and windowing (Default: ``True``) remove_dc_offset (bool, optional): Subtract mean from waveform on each frame (Default: ``True``) round_to_power_of_two (bool, optional): If True, round window size to power of two by zero-padding input to FFT. (Default: ``True``) sample_frequency (float, optional): Waveform data sample frequency (must match the waveform file, if specified there) (Default: ``16000.0``) snip_edges (bool, optional): If True, end effects will be handled by outputting only frames that completely fit in the file, and the number of frames depends on the frame_length. If False, the number of frames depends only on the frame_shift, and we reflect the data at the ends. (Default: ``True``) subtract_mean (bool, optional): Subtract mean of each feature file [CMS]; not recommended to do it this way. (Default: ``False``) use_energy (bool, optional): Add an extra dimension with energy to the FBANK output. (Default: ``False``) use_log_fbank (bool, optional):If true, produce log-filterbank, else produce linear. (Default: ``True``) use_power (bool, optional): If true, use power, else use magnitude. (Default: ``True``) vtln_high (float, optional): High inflection point in piecewise linear VTLN warping function (if negative, offset from high-mel-freq (Default: ``-500.0``) vtln_low (float, optional): Low inflection point in piecewise linear VTLN warping function (Default: ``100.0``) vtln_warp (float, optional): Vtln warp factor (only applicable if vtln_map not specified) (Default: ``1.0``) window_type (str, optional): Type of window ('hamming'|'hanning'|'povey'|'rectangular'|'blackman') (Default: ``'povey'``) Returns: Tensor: A fbank identical to what Kaldi would output. The shape is (m, ``num_mel_bins + use_energy``) where m is calculated in _get_strided rrrj)rrr\r]rr!)rrrUrSr%r&rfrkrlrmr5rrrcrdr*mmTrDrrFrbr'rh)$rr-rIrWrBrLrKrzrryrirrNrVrXrMrJrrgrrrrrrr,rrrrrTr@reZspectrumZ mel_energies_rrrr sfK    )num_cepsrrcCsFtj||d}tdt||dddf<|ddd|f}|S)NZorthorr) torchaudiordZ create_dctr8r`float)rr dct_matrixrrr_get_dct_matrixsr)rcepstral_lifterrcCs*t|}dd|ttj||S)NrOr1)r%r:sinr8r9)rrirrr_get_lifter_coeffss rg6@ )rr-rrIrWrBrLrKrzrryrrirrNrVrXrMrJrrgrrrrr,rc"Cs`| | kstd| | f|j|j}}t||||||||| | | | ||||||d|dd||||d}|r|dd| rv| ndf}t| }|dd|| |f}t| | j||d}||}|dkrt| | d} || j||d 9}|r||dddf<| rR|dddf d }!|ddd df}|s@|!t d 9}!t j ||!fd d }t||}|S) aMCreate a mfcc from a raw audio signal. This matches the input/output of Kaldi's compute-mfcc-feats. Args: waveform (Tensor): Tensor of audio of size (c, n) where c is in the range [0,2) blackman_coeff (float, optional): Constant coefficient for generalized Blackman window. (Default: ``0.42``) cepstral_lifter (float, optional): Constant that controls scaling of MFCCs (Default: ``22.0``) channel (int, optional): Channel to extract (-1 -> expect mono, 0 -> left, 1 -> right) (Default: ``-1``) dither (float, optional): Dithering constant (0.0 means no dither). If you turn this off, you should set the energy_floor option, e.g. to 1.0 or 0.1 (Default: ``0.0``) energy_floor (float, optional): Floor on energy (absolute, not relative) in Spectrogram computation. Caution: this floor is applied to the zeroth component, representing the total signal energy. The floor on the individual spectrogram elements is fixed at std::numeric_limits::epsilon(). (Default: ``1.0``) frame_length (float, optional): Frame length in milliseconds (Default: ``25.0``) frame_shift (float, optional): Frame shift in milliseconds (Default: ``10.0``) high_freq (float, optional): High cutoff frequency for mel bins (if <= 0, offset from Nyquist) (Default: ``0.0``) htk_compat (bool, optional): If true, put energy last. Warning: not sufficient to get HTK compatible features (need to change other parameters). (Default: ``False``) low_freq (float, optional): Low cutoff frequency for mel bins (Default: ``20.0``) num_ceps (int, optional): Number of cepstra in MFCC computation (including C0) (Default: ``13``) min_duration (float, optional): Minimum duration of segments to process (in seconds). (Default: ``0.0``) num_mel_bins (int, optional): Number of triangular mel-frequency bins (Default: ``23``) preemphasis_coefficient (float, optional): Coefficient for use in signal preemphasis (Default: ``0.97``) raw_energy (bool, optional): If True, compute energy before preemphasis and windowing (Default: ``True``) remove_dc_offset (bool, optional): Subtract mean from waveform on each frame (Default: ``True``) round_to_power_of_two (bool, optional): If True, round window size to power of two by zero-padding input to FFT. (Default: ``True``) sample_frequency (float, optional): Waveform data sample frequency (must match the waveform file, if specified there) (Default: ``16000.0``) snip_edges (bool, optional): If True, end effects will be handled by outputting only frames that completely fit in the file, and the number of frames depends on the frame_length. If False, the number of frames depends only on the frame_shift, and we reflect the data at the ends. (Default: ``True``) subtract_mean (bool, optional): Subtract mean of each feature file [CMS]; not recommended to do it this way. (Default: ``False``) use_energy (bool, optional): Add an extra dimension with energy to the FBANK output. (Default: ``False``) vtln_high (float, optional): High inflection point in piecewise linear VTLN warping function (if negative, offset from high-mel-freq (Default: ``-500.0``) vtln_low (float, optional): Low inflection point in piecewise linear VTLN warping function (Default: ``100.0``) vtln_warp (float, optional): Vtln warp factor (only applicable if vtln_map not specified) (Default: ``1.0``) window_type (str, optional): Type of window ('hamming'|'hanning'|'povey'|'rectangular'|'blackman') (Default: ``"povey"``) Returns: Tensor: A mfcc identical to what Kaldi would output. The shape is (m, ``num_ceps``) where m is calculated in _get_strided z5num_ceps cannot be larger than num_mel_bins: %d vs %dFT)rr-rIrWrBrLrKrzrryrirrNrVrXrMrJrrgrrrrrrr,Nrr rCrrrr!)r#rrr rQrrmatmulrrbr8r`r%r'rh)"rr-rrIrWrBrLrKrzrryrrirrNrVrXrMrJrrgrrrrr,rrZfeaturereZ mel_offsetrZ lifter_coeffsZenergyrrrr sbK   )+r8typingrr%rr__all__rGZfinforZepsrrRr3r2r4r6r7WINDOWSrrQrboolr+strrr?rHrUrfrhr rrrrr r rr rrr rrrrs  +    D  Y L  P