U 5dC'@sdZddlZddlZddlZddlZddlmZddl m Z ddl m Z ddgZ e ddd ejd d d d dZe d dd dddZdS)z3Harmonic calculations for frequency representationsN)ParameterError) is_unique)deprecate_positional_argssalienceinterp_harmonicsTZlinear)weights aggregate filter_peaks fill_valuekindaxisc Cs|dkrtj}|dkr(tt|f}ntj|td}t|||||d} |tjkrf|| |d|d} n|| |dd} |rtjj ||d} t |j } | || | | | <| } | S)a/ Harmonic salience function. Parameters ---------- S : np.ndarray [shape=(..., d, n)] input time frequency magnitude representation (e.g. STFT or CQT magnitudes). Must be real-valued and non-negative. freqs : np.ndarray, shape=(S.shape[axis]) The frequency values corresponding to S's elements along the chosen axis. harmonics : list-like, non-negative Harmonics to include in salience computation. The first harmonic (1) corresponds to ``S`` itself. Values less than one (e.g., 1/2) correspond to sub-harmonics. weights : list-like The weight to apply to each harmonic in the summation. (default: uniform weights). Must be the same length as ``harmonics``. aggregate : function aggregation function (default: `np.average`) If ``aggregate=np.average``, then a weighted average is computed per-harmonic according to the specified weights. For all other aggregation functions, all harmonics are treated equally. filter_peaks : bool If true, returns harmonic summation only on frequencies of peak magnitude. Otherwise returns harmonic summation over the full spectrum. Defaults to True. fill_value : float The value to fill non-peaks in the output representation. (default: `np.nan`) Only used if ``filter_peaks == True``. kind : str Interpolation type for harmonic estimation. See `scipy.interpolate.interp1d`. axis : int The axis along which to compute harmonics Returns ------- S_sal : np.ndarray ``S_sal`` will have the same shape as ``S``, and measure the overall harmonic energy at each frequency. See Also -------- interp_harmonics Examples -------- >>> y, sr = librosa.load(librosa.ex('trumpet'), duration=3) >>> S = np.abs(librosa.stft(y)) >>> freqs = librosa.fft_frequencies(sr=sr) >>> harms = [1, 2, 3, 4] >>> weights = [1.0, 0.5, 0.33, 0.25] >>> S_sal = librosa.salience(S, freqs=freqs, harmonics=harms, weights=weights, fill_value=0) >>> print(S_sal.shape) (1025, 115) >>> import matplotlib.pyplot as plt >>> fig, ax = plt.subplots(nrows=2, sharex=True, sharey=True) >>> librosa.display.specshow(librosa.amplitude_to_db(S, ref=np.max), ... sr=sr, y_axis='log', x_axis='time', ax=ax[0]) >>> ax[0].set(title='Magnitude spectrogram') >>> ax[0].label_outer() >>> img = librosa.display.specshow(librosa.amplitude_to_db(S_sal, ... ref=np.max), ... sr=sr, y_axis='log', x_axis='time', ax=ax[1]) >>> ax[1].set(title='Salience spectrogram') >>> fig.colorbar(img, ax=ax, format="%+2.0f dB") N)Zdtype)freqs harmonicsr r)rr r) npZaverageZoneslenarrayfloatrscipysignalZ argrelmaxemptyshapefill) Srrr r r r r rZS_harmZS_salZS_peaksZS_outr9/tmp/pip-unpacked-wheel-8l90aumz/librosa/core/harmonic.pyrs Z    )r r rc s|jdkrft||j|krft|dds6tjdddtjj|||ddd}t j |}||S|j|jkrt t||dstjdddfd d }t j |d d } | ||d ||d d|d |dStd|j|jdS)a: Compute the energy at harmonics of time-frequency representation. Given a frequency-based energy representation such as a spectrogram or tempogram, this function computes the energy at the chosen harmonics of the frequency axis. (See examples below.) The resulting harmonic array can then be used as input to a salience computation. Parameters ---------- x : np.ndarray The input energy freqs : np.ndarray, shape=(X.shape[axis]) The frequency values corresponding to X's elements along the chosen axis. harmonics : list-like, non-negative Harmonics to compute as ``harmonics[i] * freqs``. The first harmonic (1) corresponds to ``freqs``. Values less than one (e.g., 1/2) correspond to sub-harmonics. kind : str Interpolation type. See `scipy.interpolate.interp1d`. fill_value : float The value to fill when extrapolating beyond the observed frequency range. axis : int The axis along which to compute harmonics Returns ------- x_harm : np.ndarray ``x_harm[i]`` will have the same shape as ``x``, and measure the energy at the ``harmonics[i]`` harmonic of each frequency. A new dimension indexing harmonics will be inserted immediately before ``axis``. See Also -------- scipy.interpolate.interp1d Examples -------- Estimate the harmonics of a time-averaged tempogram >>> y, sr = librosa.load(librosa.ex('sweetwaltz')) >>> # Compute the time-varying tempogram and average over time >>> tempi = np.mean(librosa.feature.tempogram(y=y, sr=sr), axis=1) >>> # We'll measure the first five harmonics >>> harmonics = [1, 2, 3, 4, 5] >>> f_tempo = librosa.tempo_frequencies(len(tempi), sr=sr) >>> # Build the harmonic tensor; we only have one axis here (tempo) >>> t_harmonics = librosa.interp_harmonics(tempi, freqs=f_tempo, harmonics=harmonics, axis=0) >>> print(t_harmonics.shape) (5, 384) >>> # And plot the results >>> import matplotlib.pyplot as plt >>> fig, ax = plt.subplots() >>> librosa.display.specshow(t_harmonics, x_axis='tempo', sr=sr, ax=ax) >>> ax.set(yticks=np.arange(len(harmonics)), ... yticklabels=['{:.3g}'.format(_) for _ in harmonics], ... ylabel='Harmonic', xlabel='Tempo (BPM)') We can also compute frequency harmonics for spectrograms. To calculate sub-harmonic energy, use values < 1. >>> y, sr = librosa.load(librosa.ex('trumpet'), duration=3) >>> harmonics = [1./3, 1./2, 1, 2, 3, 4] >>> S = np.abs(librosa.stft(y)) >>> fft_freqs = librosa.fft_frequencies(sr=sr) >>> S_harm = librosa.interp_harmonics(S, freqs=fft_freqs, harmonics=harmonics, axis=0) >>> print(S_harm.shape) (6, 1025, 646) >>> fig, ax = plt.subplots(nrows=3, ncols=2, sharex=True, sharey=True) >>> for i, _sh in enumerate(S_harm): ... img = librosa.display.specshow(librosa.amplitude_to_db(_sh, ... ref=S.max()), ... sr=sr, y_axis='log', x_axis='time', ... ax=ax.flat[i]) ... ax.flat[i].set(title='h={:.3g}'.format(harmonics[i])) ... ax.flat[i].label_outer() >>> fig.colorbar(img, ax=ax, format="%+2.f dB") rrrzOFrequencies are not unique. This may produce incorrect harmonic interpolations.r) stacklevelF)r bounds_errorcopyr r cs*tjj||ddd}|tj|S)NF)r r!r r )r interpolateinterp1drmultiplyouter)Z_a_bZinterpr rr rr _f_interpsz#interp_harmonics.._f_interpz(f),(f)->(f,h)) signaturerz,freqs.shape={} does not match input shape={}N)ndimrrrwarningswarnrr"r#rr$r%allZ vectorizeZswapaxesrformat) xrrr r rZf_interpZf_outr(Zxfuncrr'rrsNV    )__doc__r,ZnumpyrZscipy.interpolaterZ scipy.signalZutil.exceptionsrutilrZutil.decoratorsr__all__nanrrrrrrs$   s