U 5d% @sdZddlZddlZddlZddlmZddlm Z ddlm Z ddl m Z ddl mZdd lmZmZdd lmZd d d dgZedddddd Zeddddddddddejd dd Zedddddddd ZeddddddddZdS) zFeature inversionN)ParameterError) griffinlim) db_to_power)tiny)filters)nnls expand_to)deprecate_positional_args mel_to_stft mel_to_audio mfcc_to_mel mfcc_to_audioi"Vig@srn_fftpowercKs@tjf|||jd|jd|}t||}tj|d||dS)aApproximate STFT magnitude from a Mel power spectrogram. Parameters ---------- M : np.ndarray [shape=(..., n_mels, n), non-negative] The spectrogram as produced by `feature.melspectrogram` sr : number > 0 [scalar] sampling rate of the underlying signal n_fft : int > 0 [scalar] number of FFT components in the resulting STFT power : float > 0 [scalar] Exponent for the magnitude melspectrogram **kwargs : additional keyword arguments Mel filter bank parameters. See `librosa.filters.mel` for details Returns ------- S : np.ndarray [shape=(..., n_fft, t), non-negative] An approximate linear magnitude spectrogram See Also -------- librosa.feature.melspectrogram librosa.stft librosa.filters.mel librosa.util.nnls Examples -------- >>> y, sr = librosa.load(librosa.ex('trumpet')) >>> S = np.abs(librosa.stft(y)) >>> mel_spec = librosa.feature.melspectrogram(S=S, sr=sr) >>> S_inv = librosa.feature.inverse.mel_to_stft(mel_spec, sr=sr) Compare the results visually >>> import matplotlib.pyplot as plt >>> fig, ax = plt.subplots(nrows=3, sharex=True, sharey=True) >>> img = librosa.display.specshow(librosa.amplitude_to_db(S, ref=np.max, top_db=None), ... y_axis='log', x_axis='time', ax=ax[0]) >>> ax[0].set(title='Original STFT') >>> ax[0].label_outer() >>> librosa.display.specshow(librosa.amplitude_to_db(S_inv, ref=np.max, top_db=None), ... y_axis='log', x_axis='time', ax=ax[1]) >>> ax[1].set(title='Reconstructed STFT') >>> ax[1].label_outer() >>> librosa.display.specshow(librosa.amplitude_to_db(np.abs(S_inv - S), ... ref=S.max(), top_db=None), ... vmax=0, y_axis='log', x_axis='time', cmap='magma', ax=ax[2]) >>> ax[2].set(title='Residual error (dB)') >>> fig.colorbar(img, ax=ax, format="%+2.f dB") )rrn_melsdtype?)out)rZmelshaperrnpr)MrrrkwargsZ mel_basisZinverser;/tmp/pip-unpacked-wheel-8l90aumz/librosa/feature/inverse.pyr s9 ZhannTZconstant ) rr hop_length win_lengthwindowcenterpad_modern_iterlengthrc  Ks4t|f|||d| } t| | |||||| | |d S)a{Invert a mel power spectrogram to audio using Griffin-Lim. This is primarily a convenience wrapper for: >>> S = librosa.feature.inverse.mel_to_stft(M) >>> y = librosa.griffinlim(S) Parameters ---------- M : np.ndarray [shape=(..., n_mels, n), non-negative] The spectrogram as produced by `feature.melspectrogram` sr : number > 0 [scalar] sampling rate of the underlying signal n_fft : int > 0 [scalar] number of FFT components in the resulting STFT hop_length : None or int > 0 The hop length of the STFT. If not provided, it will default to ``n_fft // 4`` win_length : None or int > 0 The window length of the STFT. By default, it will equal ``n_fft`` window : string, tuple, number, function, or np.ndarray [shape=(n_fft,)] A window specification as supported by `stft` or `istft` center : boolean If `True`, the STFT is assumed to use centered frames. If `False`, the STFT is assumed to use left-aligned frames. pad_mode : string If ``center=True``, the padding mode to use at the edges of the signal. By default, STFT uses zero padding. power : float > 0 [scalar] Exponent for the magnitude melspectrogram n_iter : int > 0 The number of iterations for Griffin-Lim length : None or int > 0 If provided, the output ``y`` is zero-padded or clipped to exactly ``length`` samples. dtype : np.dtype Real numeric type for the time-domain signal. Default is 32-bit float. **kwargs : additional keyword arguments Mel filter bank parameters Returns ------- y : np.ndarray [shape(..., n,)] time-domain signal reconstructed from ``M`` See Also -------- librosa.griffinlim librosa.feature.melspectrogram librosa.filters.mel librosa.feature.inverse.mel_to_stft r) r$rr rr!r"rr%r#)r r)rrrrr r!r"r#rr$r%rrZstftrrrr YsEZorthorrdct_typenormreflifterc Cs|dkr|jd}tjdd||jd}t||jdd}d|dttj||}tt |t |jj krt j dtdd ||t|}n|dkrtd tjj|d|||d } t| |d S) aoInvert Mel-frequency cepstral coefficients to approximate a Mel power spectrogram. This inversion proceeds in two steps: 1. The inverse DCT is applied to the MFCCs 2. `librosa.db_to_power` is applied to map the dB-scaled result to a power spectrogram Parameters ---------- mfcc : np.ndarray [shape=(..., n_mfcc, n)] The Mel-frequency cepstral coefficients n_mels : int > 0 The number of Mel frequencies dct_type : {1, 2, 3} Discrete cosine transform (DCT) type By default, DCT type-2 is used. norm : None or 'ortho' If ``dct_type`` is `2 or 3`, setting ``norm='ortho'`` uses an orthonormal DCT basis. Normalization is not supported for `dct_type=1`. ref : number or callable Reference power for (inverse) decibel calculation lifter : number >= 0 If ``lifter>0``, apply inverse liftering (inverse cepstral filtering):: M[n, :] <- M[n, :] / (1 + sin(pi * (n + 1) / lifter) * lifter / 2) Returns ------- M : np.ndarray [shape=(..., n_mels, n)] An approximate Mel power spectrum recovered from ``mfcc`` Warns ----- UserWarning due to critical values in lifter array that invokes underflow. See Also -------- librosa.feature.mfcc librosa.feature.melspectrogram scipy.fftpack.dct rr)r)ndimZaxesg?z@lifter array includes critical values that may invoke underflow.r)messagecategory stacklevelz1MFCC to mel lifter must be a non-negative number.)Zaxistyper)n)r*)rrZarangerr r-sinpianyabsZfinfoZepswarningswarn UserWarningrrscipyZfftpackZidctr) mfccrr(r)r*r+Zn_mfccidxZ lifter_sineZlogmelrrrr s 4  cKs t||||||d}t|f|S)axConvert Mel-frequency cepstral coefficients to a time-domain audio signal This function is primarily a convenience wrapper for the following steps: 1. Convert mfcc to Mel power spectrum (`mfcc_to_mel`) 2. Convert Mel power spectrum to time-domain audio (`mel_to_audio`) Parameters ---------- mfcc : np.ndarray [shape=(..., n_mfcc, n)] The Mel-frequency cepstral coefficients n_mels : int > 0 The number of Mel frequencies dct_type : {1, 2, 3} Discrete cosine transform (DCT) type By default, DCT type-2 is used. norm : None or 'ortho' If ``dct_type`` is `2 or 3`, setting ``norm='ortho'`` uses an orthonormal DCT basis. Normalization is not supported for ``dct_type=1``. ref : number or callable Reference power for (inverse) decibel calculation lifter : number >= 0 If ``lifter>0``, apply inverse liftering (inverse cepstral filtering):: M[n, :] <- M[n, :] / (1 + sin(pi * (n + 1) / lifter)) * lifter / 2 **kwargs : additional keyword arguments Parameters to pass through to `mel_to_audio` Returns ------- y : np.ndarray [shape=(..., n)] A time-domain signal reconstructed from `mfcc` See Also -------- mfcc_to_mel mel_to_audio librosa.feature.mfcc librosa.griffinlim scipy.fftpack.dct r')r r )r;rr(r)r*r+rZmel_specrrrrs5)__doc__r7ZnumpyrZ scipy.fftpackr:Zutil.exceptionsrZ core.spectrumrrZ util.utilsrrutilrr Zutil.decoratorsr __all__r Zfloat32r r rrrrrsF       CTK