}Kg#(.dZddlZddlZddlZddlZddlmZddlm Z ddl m Z ddlm Z dd lm Z dd lmZdd lmZdd lmZmZmZmZmZdd lmZmZmZgdZddddddddeej:dedeej:deededededefdZ ddddedededefdZ!e d !dddddd"d#d$dd%d&d'dd( deej:dedeej:deed)eed*ed+ed,ed-eed.ed/e"d0ed1eeeefdeej:ej:ffd2Z#d3ej:d4ed-ed5ed6edej:f d7Z$ejJd8ej:dej:fd9Z&ejNd:d;gdZ(d?d@d8ej:dAedej:fdBZ)ddddd$d&d'dCdej:d*ed+eded4ed-eed)eedDed/e"d0edej:fdEZ*dddddFdGdd$dHddejVd&d'dIdej:d*ed+eded4ed-eed)eedJedKeeefdLededMedNedOedPeed/e"d0edeej:ej:ej:ff$dQZ,dRZ-ded+ed*ed4ed-ef dSZ.y)Tz$Pitch-tracking and tuning estimationN) _spectrogram)convert)cache)util)sequence)ParameterError) ArrayLike)AnyCallableOptionalTupleUnion) _WindowSpec_PadMode _PadModeSTFT)estimate_tuning pitch_tuningpiptrackyinpyini"Vig{Gz? )ysrSn_fft resolutionbins_per_octaverrrrrrkwargsreturnc td||||d|\}}|dkD} | jrtj|| } nd} t ||| k\| z||S)aEstimate the tuning of an audio time series or spectrogram input. Parameters ---------- y : np.ndarray [shape=(..., n)] or None audio signal. Multi-channel is supported.. sr : number > 0 [scalar] audio sampling rate of ``y`` S : np.ndarray [shape=(..., d, t)] or None magnitude or power spectrogram n_fft : int > 0 [scalar] or None number of FFT bins to use, if ``y`` is provided. resolution : float in `(0, 1)` Resolution of the tuning as a fraction of a bin. 0.01 corresponds to measurements in cents. bins_per_octave : int > 0 [scalar] How many frequency bins per octave **kwargs : additional keyword arguments Additional arguments passed to `piptrack` Returns ------- tuning: float in `[-0.5, 0.5)` estimated tuning deviation (fractions of a bin). Note that if multichannel input is provided, a single tuning estimate is provided spanning all channels. See Also -------- piptrack : Pitch tracking by parabolic interpolation Examples -------- With time-series input >>> y, sr = librosa.load(librosa.ex('trumpet')) >>> librosa.estimate_tuning(y=y, sr=sr) -0.08000000000000002 In tenths of a cent >>> librosa.estimate_tuning(y=y, sr=sr, resolution=1e-3) -0.016000000000000014 Using spectrogram input >>> S = np.abs(librosa.stft(y)) >>> librosa.estimate_tuning(S=S, sr=sr) -0.08000000000000002 Using pass-through arguments to `librosa.piptrack` >>> librosa.estimate_tuning(y=y, sr=sr, n_fft=8192, ... fmax=librosa.note_to_hz('G#9')) -0.08000000000000002 )rrrrrrr)ranynpmedianr) rrrrrrr pitchmag pitch_mask thresholds V/home/alanp/www/video.onchill/myenv/lib/python3.12/site-packages/librosa/core/pitch.pyrrsrFAA"A&AJE3J~~IIc*o.   si:-.' r$ frequenciesc tj|}||dkD}tj|stjddytj |t j|zd}||dk\xxdzcc<tjddttjd|z d z}tj||\}}|tj|}|S) aGiven a collection of pitches, estimate its tuning offset (in fractions of a bin) relative to A440=440.0Hz. Parameters ---------- frequencies : array-like, float A collection of frequencies detected in the signal. See `piptrack` resolution : float in `(0, 1)` Resolution of the tuning as a fraction of a bin. 0.01 corresponds to cents. bins_per_octave : int > 0 [scalar] How many frequency bins per octave Returns ------- tuning: float in `[-0.5, 0.5)` estimated tuning deviation (fractions of a bin) See Also -------- estimate_tuning : Estimating tuning from time-series or spectrogram input Examples -------- >>> # Generate notes at +25 cents >>> freqs = librosa.cqt_frequencies(n_bins=24, fmin=55, tuning=0.25) >>> librosa.pitch_tuning(freqs) 0.25 >>> # Track frequencies from a real spectrogram >>> y, sr = librosa.load(librosa.ex('trumpet')) >>> freqs, times, mags = librosa.reassigned_spectrogram(y, sr=sr, ... fill_nan=True) >>> # Select out pitches with high energy >>> freqs = freqs[mags > np.median(mags)] >>> librosa.pitch_tuning(freqs) -0.07 rz3Trying to estimate tuning from empty frequency set.r) stacklevelr#??gr) r' atleast_1dr&warningswarnmodr hz_to_octslinspaceintceil histogramargmax)r/rrresidualbinscountstuning tuning_ests r-rrlsT-- ,KkAo.K 66+  Aa vvo(:(:;(GGMH  X_$ ;;tS#bggcJ.>&?"@1"D ED\\(D1NFFryy01J r.)levelgb@g@@g?hannTconstant) rrrr hop_lengthfminfmaxr, win_lengthwindowcenterpad_moderefrGrHrIr,rJrKrLrMrNc lt|||||| | | \}}tj|}tj|d}tj|t |dz }t j||} tj|d}t|d}d|z|z}tj|}tj|}|| k| |kz}tj||jd}| tj} t| r$|| |dz}tj |d}ntj| }tj"|tj$|||kDzdz}|d||zt |z|z ||<||||z||<||fS) a Pitch tracking on thresholded parabolically-interpolated STFT. This implementation uses the parabolic interpolation method described by [#]_. .. [#] https://ccrma.stanford.edu/~jos/sasp/Sinusoidal_Peak_Interpolation.html Parameters ---------- y : np.ndarray [shape=(..., n)] or None audio signal. Multi-channel is supported.. sr : number > 0 [scalar] audio sampling rate of ``y`` S : np.ndarray [shape=(..., d, t)] or None magnitude or power spectrogram n_fft : int > 0 [scalar] or None number of FFT bins to use, if ``y`` is provided. hop_length : int > 0 [scalar] or None number of samples to hop threshold : float in `(0, 1)` A bin in spectrum ``S`` is considered a pitch when it is greater than ``threshold * ref(S)``. By default, ``ref(S)`` is taken to be ``max(S, axis=0)`` (the maximum value in each column). fmin : float > 0 [scalar] lower frequency cutoff. fmax : float > 0 [scalar] upper frequency cutoff. win_length : int <= n_fft [scalar] Each frame of audio is windowed by ``window``. The window will be of length `win_length` and then padded with zeros to match ``n_fft``. If unspecified, defaults to ``win_length = n_fft``. window : string, tuple, number, function, or np.ndarray [shape=(n_fft,)] - a window specification (string, tuple, or number); see `scipy.signal.get_window` - a window function, such as `scipy.signal.windows.hann` - a vector or array of length ``n_fft`` .. see also:: `filters.get_window` center : boolean - If ``True``, the signal ``y`` is padded so that frame ``t`` is centered at ``y[t * hop_length]``. - If ``False``, then frame ``t`` begins at ``y[t * hop_length]`` pad_mode : string If ``center=True``, the padding mode to use at the edges of the signal. By default, STFT uses zero-padding. See also: `np.pad`. ref : scalar or callable [default=np.max] If scalar, the reference value against which ``S`` is compared for determining pitches. If callable, the reference value is computed as ``ref(S, axis=0)``. Returns ------- pitches, magnitudes : np.ndarray [shape=(..., d, t)] Where ``d`` is the subset of FFT bins within ``fmin`` and ``fmax``. ``pitches[..., f, t]`` contains instantaneous frequency at bin ``f``, time ``t`` ``magnitudes[..., f, t]`` contains the corresponding magnitudes. Both ``pitches`` and ``magnitudes`` take value 0 at bins of non-maximal magnitude. Notes ----- This function caches at level 30. One of ``S`` or ``y`` must be provided. If ``S`` is not given, it is computed from ``y`` using the default parameters of `librosa.stft`. Examples -------- Computing pitches from a waveform input >>> y, sr = librosa.load(librosa.ex('trumpet')) >>> pitches, magnitudes = librosa.piptrack(y=y, sr=sr) Or from a spectrogram input >>> S = np.abs(librosa.stft(y)) >>> pitches, magnitudes = librosa.piptrack(S=S, sr=sr) Or with an alternate reference value for pitch detection, where values above the mean spectral energy in each frame are counted as pitches >>> pitches, magnitudes = librosa.piptrack(S=S, sr=sr, threshold=1, ... ref=np.mean) )rrrrGrJrKrLrMrr)rraxisr3ndimaxes)rr'absmaximumminimumfloatrfft_frequenciesgradient_parabolic_interpolation zeros_liker expand_torTmaxcallable expand_dimsnonzerolocalmax)rrrrrGrHrIr,rJrKrLrMrN fft_freqsavgshiftdskewpitchesmags freq_mask ref_valueidxs r-rrsz  HAu q A ::dA D ::dE"IM *D''2U;I ++ab !C $QR 0E #I EmmAG == D"y4'78IyqvvB?I {ff}AB/ NN9b1 FF3K  **YqA M/B!LL MCGeCj(E"I5=GCL#s#DI D=r.y_frames frame_length min_period max_periodc2tjj||d}tjj|d|ddddf|d}tjj||z|dd|dddf}d|tj|dk<tj |dzd}|d|dddf|dd| ddfz }d|tj|dk<|ddd ddf|zd|zz } | d||d zddf} t jtjd |d z| jd } tj | dd |d zddfd| z } | d|d z |ddf} | | t j| zz } | S) ugCumulative mean normalized difference function (equation 8 in [#]_) .. [#] De Cheveigné, Alain, and Hideki Kawahara. "YIN, a fundamental frequency estimator for speech and music." The Journal of the Acoustical Society of America 111.4 (2002): 1917-1930. Parameters ---------- y_frames : np.ndarray [shape=(frame_length, n_frames)] framed audio time series. frame_length : int > 0 [scalar] length of the frames in samples. win_length : int > 0 [scalar] length of the window for calculating autocorrelation in samples. min_period : int > 0 [scalar] minimum period. max_period : int > 0 [scalar] maximum period. Returns ------- yin_frames : np.ndarray [shape=(max_period-min_period+1,n_frames)] Cumulative mean normalized difference function for each frame. rPrQ.rNgư>rrrS) r'fftrfftirfftrVcumsumrr^arangerTtiny)rmrnrJrorpab acf_frames energy_frames yin_frames yin_numerator tau_rangecumulative_meanyin_denominators r-&_cumulative_mean_normalized_differenceris@  Hl 4A  HS*Qr/145|" MAa!e\;CaJ+F&IJO*$))O44J r.xc|d|dzd|dzz }|d|dz dz }tj|tj|k\ry| |z S)z0Stencil to compute local parabolic interpolationrrrrr)r'rV)rryrzs r- _pi_stencilrs` !qu q1Q4xA 1"A vvayBFF1I 26Mr.zvoid(float32[:], float32[:])zvoid(float64[:], float64[:])z(n)->(n))rnopythonc t||ddy)z:Vectorized wrapper for the parabolic interpolation stencilN)r)rrs r- _pi_wrapperrs q>AaDr.rPrQrRc|jd|}tj|}|jd|}t||d|d<d|d<|S)aPiecewise parabolic interpolation for yin and pyin. Parameters ---------- x : np.ndarray array to interpolate axis : int axis along which to interpolate Returns ------- parabolic_shifts : np.ndarray [shape=x.shape] position of the parabola optima (relative to bin indices) Note: the shift at bin `n` is determined as 0 if the estimated optimum is outside the range `[n-1, n+1]`. rrr).rr).r)swapaxesr' empty_liker)rrRxishiftsshiftsis r-r\r\sZ& B B]]1 Foob$'GGGGGFO Mr.)rrnrJrGtrough_thresholdrLrMrc  || td||dz}t|||||||dz}tj|d|r5dg|jz} |dz|dzf| d <t j || | }tj||| } tt j||z } ttt j||z ||z d z } t| ||| | }t|}tj|d }|ddddf|dd ddfk|ddddf<t j|||k}t!|j"}d |d <t j$|d }t j&|d }|j)|}|j)|}t j*|d d}||||<| |zt j,||d zddddf}||z }|S)u Fundamental frequency (F0) estimation using the YIN algorithm. YIN is an autocorrelation based method for fundamental frequency estimation [#]_. First, a normalized difference function is computed over short (overlapping) frames of audio. Next, the first minimum in the difference function below ``trough_threshold`` is selected as an estimate of the signal's period. Finally, the estimated period is refined using parabolic interpolation before converting into the corresponding frequency. .. [#] De Cheveigné, Alain, and Hideki Kawahara. "YIN, a fundamental frequency estimator for speech and music." The Journal of the Acoustical Society of America 111.4 (2002): 1917-1930. Parameters ---------- y : np.ndarray [shape=(..., n)] audio time series. Multi-channel is supported.. fmin : number > 0 [scalar] minimum frequency in Hertz. The recommended minimum is ``librosa.note_to_hz('C2')`` (~65 Hz) though lower values may be feasible. fmax : number > fmin, <= sr/2 [scalar] maximum frequency in Hertz. The recommended maximum is ``librosa.note_to_hz('C7')`` (~2093 Hz) though higher values may be feasible. sr : number > 0 [scalar] sampling rate of ``y`` in Hertz. frame_length : int > 0 [scalar] length of the frames in samples. By default, ``frame_length=2048`` corresponds to a time scale of about 93 ms at a sampling rate of 22050 Hz. win_length : None or int > 0 [scalar] length of the window for calculating autocorrelation in samples. If ``None``, defaults to ``frame_length // 2`` hop_length : None or int > 0 [scalar] number of audio samples between adjacent YIN predictions. If ``None``, defaults to ``frame_length // 4``. trough_threshold : number > 0 [scalar] absolute threshold for peak estimation. center : boolean If ``True``, the signal `y` is padded so that frame ``D[:, t]`` is centered at `y[t * hop_length]`. If ``False``, then ``D[:, t]`` begins at ``y[t * hop_length]``. Defaults to ``True``, which simplifies the alignment of ``D`` onto a time grid by means of ``librosa.core.frames_to_samples``. pad_mode : string or function If ``center=True``, this argument is passed to ``np.pad`` for padding the edges of the signal ``y``. By default (``pad_mode="constant"``), ``y`` is padded on both sides with zeros. If ``center=False``, this argument is ignored. .. see also:: `np.pad` Returns ------- f0: np.ndarray [shape=(..., n_frames)] time series of fundamental frequencies in Hertz. If multi-channel input is provided, f0 curves are estimated separately for each channel. See Also -------- librosa.pyin : Fundamental frequency (F0) estimation using probabilistic YIN (pYIN). Examples -------- Computing a fundamental frequency (F0) curve from an audio input >>> y = librosa.chirp(fmin=440, fmax=880, duration=5.0, sr=22050) >>> librosa.yin(y, fmin=440, fmax=880, sr=22050) array([442.66354675, 441.95299983, 441.58010963, ..., 871.161732 , 873.99001454, 877.04297681]) N'both "fmin" and "fmax" must be providedrrrIrHrnrJFmonorrrrmodernrGrrPrQ.rTrRkeepdims)r __check_yin_paramsr valid_audiorTr'padframer:floorminr;rr\localmin logical_andlistshapeargminr=reshapealltake_along_axis)rrHrIrrnrJrGrrLrMpaddingrmrorpr}parabolic_shifts is_troughis_threshold_trough target_shape global_min yin_periodno_trough_below_thresholdf0s r-rrs>l |t|FGG!Q&  Dt,: !Q&  QU#(QVV##q(,!*;< FF1gH -zz!,:NHRXXb4i()JSd+,lZ.G!.KLJ8, J J 0 ; jr2I%c1ai0:c1ai3HHIc1ai..JAQ4QR  (()LL:B/J.R8J##L1J##L1J "(;';"t T,67P,QJ()     -z C D 1ai J*_B Ir.d)rg(\A@)rrnrJrG n_thresholdsbeta_parametersboltzmann_parameterrmax_transition_rate switch_probno_trough_probfill_narLrMrrrrrrrc  %&'()| td||dz}t|||||dz}tj|d|r>|jDcgc]}d}}|dz|dzf|d <t j ||| }tj||| }tt j|z &ttt jz ||z d z }t|||&|}t|}t jd d |d z)tj j"j%)|d |d }t j&|%tt jd| z 'tt jd'zt j(|z zd z(% &'( )f d}t j*|d}|||\}}t-| dz|zz }|'zd z}t/j0(|dd}t/j2dd | z }t j4||}t j6d(z} d (z | (dt/j8||| }!dt j:(d'zz zz}"|"|!(z}#|!(k}$|||#|$<|#dd ddf|$dd ddf|dd ddffScc}w)uMFundamental frequency (F0) estimation using probabilistic YIN (pYIN). pYIN [#]_ is a modificatin of the YIN algorithm [#]_ for fundamental frequency (F0) estimation. In the first step of pYIN, F0 candidates and their probabilities are computed using the YIN algorithm. In the second step, Viterbi decoding is used to estimate the most likely F0 sequence and voicing flags. .. [#] Mauch, Matthias, and Simon Dixon. "pYIN: A fundamental frequency estimator using probabilistic threshold distributions." 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2014. .. [#] De Cheveigné, Alain, and Hideki Kawahara. "YIN, a fundamental frequency estimator for speech and music." The Journal of the Acoustical Society of America 111.4 (2002): 1917-1930. Parameters ---------- y : np.ndarray [shape=(..., n)] audio time series. Multi-channel is supported. fmin : number > 0 [scalar] minimum frequency in Hertz. The recommended minimum is ``librosa.note_to_hz('C2')`` (~65 Hz) though lower values may be feasible. fmax : number > fmin, <= sr/2 [scalar] maximum frequency in Hertz. The recommended maximum is ``librosa.note_to_hz('C7')`` (~2093 Hz) though higher values may be feasible. sr : number > 0 [scalar] sampling rate of ``y`` in Hertz. frame_length : int > 0 [scalar] length of the frames in samples. By default, ``frame_length=2048`` corresponds to a time scale of about 93 ms at a sampling rate of 22050 Hz. win_length : None or int > 0 [scalar] length of the window for calculating autocorrelation in samples. If ``None``, defaults to ``frame_length // 2`` hop_length : None or int > 0 [scalar] number of audio samples between adjacent pYIN predictions. If ``None``, defaults to ``frame_length // 4``. n_thresholds : int > 0 [scalar] number of thresholds for peak estimation. beta_parameters : tuple shape parameters for the beta distribution prior over thresholds. boltzmann_parameter : number > 0 [scalar] shape parameter for the Boltzmann distribution prior over troughs. Larger values will assign more mass to smaller periods. resolution : float in `(0, 1)` Resolution of the pitch bins. 0.01 corresponds to cents. max_transition_rate : float > 0 maximum pitch transition rate in octaves per second. switch_prob : float in ``(0, 1)`` probability of switching from voiced to unvoiced or vice versa. no_trough_prob : float in ``(0, 1)`` maximum probability to add to global minimum if no trough is below threshold. fill_na : None, float, or ``np.nan`` default value for unvoiced frames of ``f0``. If ``None``, the unvoiced frames will contain a best guess value. center : boolean If ``True``, the signal ``y`` is padded so that frame ``D[:, t]`` is centered at ``y[t * hop_length]``. If ``False``, then ``D[:, t]`` begins at ``y[t * hop_length]``. Defaults to ``True``, which simplifies the alignment of ``D`` onto a time grid by means of ``librosa.core.frames_to_samples``. pad_mode : string or function If ``center=True``, this argument is passed to ``np.pad`` for padding the edges of the signal ``y``. By default (``pad_mode="constant"``), ``y`` is padded on both sides with zeros. If ``center=False``, this argument is ignored. .. see also:: `np.pad` Returns ------- f0: np.ndarray [shape=(..., n_frames)] time series of fundamental frequencies in Hertz. voiced_flag: np.ndarray [shape=(..., n_frames)] time series containing boolean flags indicating whether a frame is voiced or not. voiced_prob: np.ndarray [shape=(..., n_frames)] time series containing the probability that a frame is voiced. .. note:: If multi-channel input is provided, f0 and voicing are estimated separately for each channel. See Also -------- librosa.yin : Fundamental frequency (F0) estimation using the YIN algorithm. Examples -------- Computing a fundamental frequency (F0) curve from an audio input >>> y, sr = librosa.load(librosa.ex('trumpet')) >>> f0, voiced_flag, voiced_probs = librosa.pyin(y, ... sr=sr, ... fmin=librosa.note_to_hz('C2'), ... fmax=librosa.note_to_hz('C7')) >>> times = librosa.times_like(f0, sr=sr) Overlay F0 over a spectrogram >>> import matplotlib.pyplot as plt >>> D = librosa.amplitude_to_db(np.abs(librosa.stft(y)), ref=np.max) >>> fig, ax = plt.subplots() >>> img = librosa.display.specshow(D, x_axis='time', y_axis='log', ax=ax) >>> ax.set(title='pYIN fundamental frequency estimation') >>> fig.colorbar(img, ax=ax, format="%+2.f dB") >>> ax.plot(times, f0, label='f0', color='cyan', linewidth=3) >>> ax.legend(loc='upper right') NrrrrFrrrrrrrrr2rc . t||  S)N) __pyin_helper) ryrz beta_probsrrHron_bins_per_semitone n_pitch_binsrr thresholdss r-_helperzpyin.._helper7s1           r.z(f,t),(k,t)->(1,d,t),(j,t)) signaturetriangle)rKwrap)p_init.)r rrrrr'rrr:rrr;rr\r9scipystatsbetacdfdifflog2 vectorizeroundr transition_localtransition_loopkronzerosviterbirw)*rrHrIrrnrJrGrrrrrrrrrLrM_rrmrpr}rbeta_cdfrhelperobservation_probs voiced_probmax_semitones_per_frametransition_width transitiont_switchrstatesfreqsr voiced_flagrrorrrs* ` ` ` ` @@@@@r-rrs~ |t|FGG!Q&  Dt,: !Q&  QU##$77+7a67+#q(,!*;< FF1gH -zz!,:NHRXXb4i()JSd+,lZ.G!.KLJ8, J J 0 ; Q sr/fmax rzfmax=z.3fz! cannot exceed Nyquist frequency zfmin=z must be less than fmax=rz must be strictly positivez win_length=z must be less than frame_length=rz is too small for frame_length=z , win_length=z , and sr=z. Either increase to fmax=z or frame_length=N)r r:r'r;)rrIrHrnrJ fmax_feasibleframe_length_feasibles r-rrs7 b1f}uT#J.OPRSTPTvVWW t|uT#J.FtCjQRR qyuT#J.HIJJ\!*%El^ T  j 1$d 2lZ7!;< #BGGBtG$4z$AA$E FD:<\N-XbWcclmolpq''4S&99JK`Ja c  3r.)/__doc__r5numpyr'rnumbaspectrumrr_cacherrr util.exceptionsr numpy.typingr typingr r rrr_typingrrr__all__ndarrayrYr:rrboolrrstencilr guvectorizerr\rnanrrrr%r.r-rs+ #,"8899 H #"Q Q Q  Q C= Q  QQQ Qj48PRCC+0CJMC CLR#" $ $ ',0s s s  s C= s  s s ss s s ss %x( )s 2::rzz !"ssl=jj=== =  = ZZ =@ 2:: "**  #%CD    2::"**   <>  S "** P $ $!#^ zz^ ^  ^  ^  ^ ^ ^^ ^^ZZ^L $ $+2!"!& !vv#%\I zz\I \I  \I  \I  \I \I \I\I5%<(\I\I\I\I\I\I e_!\I" #\I$%\I& 2::rzz2:: -.'\I~J6Z   %* := KN r.