}Kg,bLdZddlZddlZddlZddlZddlmZddlm Z ddlm Z ddlm Z ddl m Z mZdd l mZdd lmZdd lmZmZdd lmZmZmZmZmZdd lmZgdZedddeZdddddddddddd deej>de deej>de!de de d e"d!eeeej>fd"eejFjHd#e%d$e"d%eeeej>fej>ffd&Z&ddddd'd(d)dd*deej>de deej>de!d+e!d,ee d-ee d"eejFjHd%ej>fd.Z'dej>d!ej>d/e de d e"d%ej>f d0Z(d1Z)ejTd2d3gd4dd56d7Z+ejTd8d9gd:dd6d;Z,ejTddd6d?Z-d@Z.ejTdAdBgdCdd6dDZ/ejTdEdFgdGdd6dHZ0y)Iz^ Beat and tempo ============== .. autosummary:: :toctree: generated/ beat_track plp N)cache)core)onset)util) tempogramfourier_tempogram)tempo)ParameterError)moved vectorize)AnyCallableOptionalTupleUnion) _FloatLike_co) beat_trackr plpzlibrosa.beat.tempoz0.10.0z1.0) moved_fromversionversion_removedi"Vig^@dTframes) ysronset_envelope hop_length start_bpm tightnesstrimbpmpriorunitssparserrrrrr r!r"r#r$r%returnc t|4| tdtj|||tj}| r(|j dk7rtd|j d|j sb| rdtjgtfStj|jdd t tj|tfS|t||||| }tj|} t!j"| |j t%| j  } t'|| t||z ||} | rjtj(| } | d k(r || fS| dk(r|t+j,| |fS| dk(r|t+j.| ||fStd| || fS)a Dynamic programming beat tracker. Beats are detected in three stages, following the method of [#]_: 1. Measure onset strength 2. Estimate tempo from onset correlation 3. Pick peaks in onset strength approximately consistent with estimated tempo .. [#] Ellis, Daniel PW. "Beat tracking by dynamic programming." Journal of New Music Research 36.1 (2007): 51-60. http://labrosa.ee.columbia.edu/projects/beattrack/ Parameters ---------- y : np.ndarray [shape=(..., n)] or None audio time series sr : number > 0 [scalar] sampling rate of ``y`` onset_envelope : np.ndarray [shape=(..., m)] or None (optional) pre-computed onset strength envelope. hop_length : int > 0 [scalar] number of audio samples between successive ``onset_envelope`` values start_bpm : float > 0 [scalar] initial guess for the tempo estimator (in beats per minute) tightness : float [scalar] tightness of beat distribution around tempo trim : bool [scalar] trim leading/trailing beats with weak onsets bpm : float [scalar] or np.ndarray [shape=(...)] (optional) If provided, use ``bpm`` as the tempo instead of estimating it from ``onsets``. If multichannel, tempo estimates can be provided for all channels. Tempo estimates may also be time-varying, in which case the shape of ``bpm`` should match that of ``onset_envelope``, i.e., one estimate provided for each frame. prior : scipy.stats.rv_continuous [optional] An optional prior distribution over tempo. If provided, ``start_bpm`` will be ignored. units : {'frames', 'samples', 'time'} The units to encode detected beat events in. By default, 'frames' are used. sparse : bool If ``True`` (default), detections are returned as an array of frames, samples, or time indices (as specified by ``units=``). If ``False``, detections are encoded as a dense boolean array where ``beats[..., n]`` is true if there's a beat at frame index ``n``. .. note:: multi-channel input is only supported when ``sparse=False``. Returns ------- tempo : float [scalar, non-negative] or np.ndarray estimated global tempo (in beats per minute) If multi-channel and ``bpm`` is not provided, a separate tempo will be returned for each channel beats : np.ndarray estimated beat event locations. If `sparse=True` (default), beat locations are given in the specified units (default is frame indices). If `sparse=False` (required for multichannel input), beat events are indicated by a boolean for each frame. .. note:: If no onset strength could be detected, beat_tracker estimates 0 BPM and returns an empty list. Raises ------ ParameterError if neither ``y`` nor ``onset_envelope`` are provided, or if ``units`` is not one of 'frames', 'samples', or 'time' See Also -------- librosa.onset.onset_strength Examples -------- Track beats using time series input >>> y, sr = librosa.load(librosa.ex('choice'), duration=10) >>> tempo, beats = librosa.beat.beat_track(y=y, sr=sr) >>> tempo 135.99917763157896 Print the frames corresponding to beats >>> beats array([ 3, 21, 40, 59, 78, 96, 116, 135, 154, 173, 192, 211, 230, 249, 268, 287, 306, 325, 344, 363]) Or print them as timestamps >>> librosa.frames_to_time(beats, sr=sr) array([0.07 , 0.488, 0.929, 1.37 , 1.811, 2.229, 2.694, 3.135, 3.576, 4.017, 4.458, 4.899, 5.341, 5.782, 6.223, 6.664, 7.105, 7.546, 7.988, 8.429]) Output beat detections as a boolean array instead of frame indices >>> tempo, beats_dense = librosa.beat.beat_track(y=y, sr=sr, sparse=False) >>> beats_dense array([False, False, False, True, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, ..., False, False, True, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False, False]) Track beats using a pre-computed onset envelope >>> onset_env = librosa.onset.onset_strength(y=y, sr=sr, ... aggregate=np.median) >>> tempo, beats = librosa.beat.beat_track(onset_envelope=onset_env, ... sr=sr) >>> tempo 135.99917763157896 >>> beats array([ 3, 21, 40, 59, 78, 96, 116, 135, 154, 173, 192, 211, 230, 249, 268, 287, 306, 325, 344, 363]) Plot the beat events against the onset strength envelope >>> import matplotlib.pyplot as plt >>> hop_length = 512 >>> fig, ax = plt.subplots(nrows=2, sharex=True) >>> times = librosa.times_like(onset_env, sr=sr, hop_length=hop_length) >>> M = librosa.feature.melspectrogram(y=y, sr=sr, hop_length=hop_length) >>> librosa.display.specshow(librosa.power_to_db(M, ref=np.max), ... y_axis='mel', x_axis='time', hop_length=hop_length, ... ax=ax[0]) >>> ax[0].label_outer() >>> ax[0].set(title='Mel spectrogram') >>> ax[1].plot(times, librosa.util.normalize(onset_env), ... label='Onset strength') >>> ax[1].vlines(times[beats], 0, 1, alpha=0.5, color='r', ... linestyle='--', label='Beats') >>> ax[1].legend() Nz$y or onset_envelope must be providedrrr aggregaterz'sparse=True (default) does not support zK-dimensional inputs. Either set sparse=False or convert the signal to mono.dtypeshaper,)rrrrr#ndimaxesrsamples)rtime)rrzInvalid unit type: )r ronset_strengthnpmedianr1anyarrayintzerosr/float zeros_likebool_tempo atleast_1dr expand_torange__beat_tracker flatnonzerorframes_to_samplesframes_to_time)rrrrrr r!r"r#r$r%_bpm bpm_expandedbeatss P/home/alanp/www/video.onchill/myenv/lib/python3.12/site-packages/librosa/beat.pyrr%sT 9 !GH H--B: .%%*F!&&'(IJK K     "C01 1HH>#7#7#>$'5':':',TYY'79L ><rZ9OQZ\` aE u% H   < i //*MN N f_,,UzbQR R #6ug!>? ? <ii,)rrrr win_length tempo_min tempo_maxr#rMrNrOc<|'tj|||tj}||||krt d|d|t ||||}t j|||} | d|d| |kddf<| d|d| |kDddf<tj| |jd } tjd tj|z} || |j| z } | jd d } d|| | k<|tj|dztj|jd d zz}t j |d||j"d} tj$| dd| } tj&| dS)aPredominant local pulse (PLP) estimation. [#]_ The PLP method analyzes the onset strength envelope in the frequency domain to find a locally stable tempo for each frame. These local periodicities are used to synthesize local half-waves, which are combined such that peaks coincide with rhythmically salient frames (e.g. onset events on a musical time grid). The local maxima of the pulse curve can be taken as estimated beat positions. This method may be preferred over the dynamic programming method of `beat_track` when the tempo is expected to vary significantly over time. Additionally, since `plp` does not require the entire signal to make predictions, it may be preferable when beat-tracking long recordings in a streaming setting. .. [#] Grosche, P., & Muller, M. (2011). "Extracting predominant local pulse information from music recordings." IEEE Transactions on Audio, Speech, and Language Processing, 19(6), 1688-1701. Parameters ---------- y : np.ndarray [shape=(..., n)] or None audio time series. Multi-channel is supported. sr : number > 0 [scalar] sampling rate of ``y`` onset_envelope : np.ndarray [shape=(..., n)] or None (optional) pre-computed onset strength envelope hop_length : int > 0 [scalar] number of audio samples between successive ``onset_envelope`` values win_length : int > 0 [scalar] number of frames to use for tempogram analysis. By default, 384 frames (at ``sr=22050`` and ``hop_length=512``) corresponds to about 8.9 seconds. tempo_min, tempo_max : numbers > 0 [scalar], optional Minimum and maximum permissible tempo values. ``tempo_max`` must be at least ``tempo_min``. Set either (or both) to `None` to disable this constraint. prior : scipy.stats.rv_continuous [optional] A prior distribution over tempo (in beats per minute). By default, a uniform prior over ``[tempo_min, tempo_max]`` is used. Returns ------- pulse : np.ndarray, shape=[(..., n)] The estimated pulse curve. Maxima correspond to rhythmically salient points of time. If input is multi-channel, one pulse curve per channel is computed. See Also -------- beat_track librosa.onset.onset_strength librosa.feature.fourier_tempogram Examples -------- Visualize the PLP compared to an onset strength envelope. Both are normalized here to make comparison easier. >>> y, sr = librosa.load(librosa.ex('brahms')) >>> onset_env = librosa.onset.onset_strength(y=y, sr=sr) >>> pulse = librosa.beat.plp(onset_envelope=onset_env, sr=sr) >>> # Or compute pulse with an alternate prior, like log-normal >>> import scipy.stats >>> prior = scipy.stats.lognorm(loc=np.log(120), scale=120, s=1) >>> pulse_lognorm = librosa.beat.plp(onset_envelope=onset_env, sr=sr, ... prior=prior) >>> melspec = librosa.feature.melspectrogram(y=y, sr=sr) >>> import matplotlib.pyplot as plt >>> fig, ax = plt.subplots(nrows=3, sharex=True) >>> librosa.display.specshow(librosa.power_to_db(melspec, ... ref=np.max), ... x_axis='time', y_axis='mel', ax=ax[0]) >>> ax[0].set(title='Mel spectrogram') >>> ax[0].label_outer() >>> ax[1].plot(librosa.times_like(onset_env), ... librosa.util.normalize(onset_env), ... label='Onset strength') >>> ax[1].plot(librosa.times_like(pulse), ... librosa.util.normalize(pulse), ... label='Predominant local pulse (PLP)') >>> ax[1].set(title='Uniform tempo prior [30, 300]') >>> ax[1].label_outer() >>> ax[2].plot(librosa.times_like(onset_env), ... librosa.util.normalize(onset_env), ... label='Onset strength') >>> ax[2].plot(librosa.times_like(pulse_lognorm), ... librosa.util.normalize(pulse_lognorm), ... label='Predominant local pulse (PLP)') >>> ax[2].set(title='Log-normal tempo prior, mean=120', xlim=[5, 20]) >>> ax[2].legend() PLP local maxima can be used as estimates of beat positions. >>> tempo, beats = librosa.beat.beat_track(onset_envelope=onset_env) >>> beats_plp = np.flatnonzero(librosa.util.localmax(pulse)) >>> import matplotlib.pyplot as plt >>> fig, ax = plt.subplots(nrows=2, sharex=True, sharey=True) >>> times = librosa.times_like(onset_env, sr=sr) >>> ax[0].plot(times, librosa.util.normalize(onset_env), ... label='Onset strength') >>> ax[0].vlines(times[beats], 0, 1, alpha=0.5, color='r', ... linestyle='--', label='Beats') >>> ax[0].legend() >>> ax[0].set(title='librosa.beat.beat_track') >>> ax[0].label_outer() >>> # Limit the plot to a 15-second window >>> times = librosa.times_like(pulse, sr=sr) >>> ax[1].plot(times, librosa.util.normalize(pulse), ... label='PLP') >>> ax[1].vlines(times[beats_plp], 0, 1, alpha=0.5, color='r', ... linestyle='--', label='PLP Beats') >>> ax[1].legend() >>> ax[1].set(title='librosa.beat.plp', xlim=[5, 20]) >>> ax[1].xaxis.set_major_formatter(librosa.display.TimeFormatter()) Nr(z tempo_max=z must be larger than tempo_min=)rrrrM)rrrMr.r0g.AT)axiskeepdims?rr-)rn_fftlengthrR)rr5r6r7r r rfourier_tempo_frequenciesrrAr1log1pabslogpdfmaxtinyistftr/clip normalize) rrrrrMrNrOr#ftgramtempo_frequenciesftmag peak_valuespulses rJrrsN--B: !69 ;Q #B9+ N  %  F66 *89s% 114589s% 1145'8v{{QST HHS266&>) *E  /00))d)3K"#F5;  dii3& T 0R)SSSF JJ1J~7K7KB7O E GGE1dE *E >>%b ))rK frame_ratectj|dkrtd|d|dkr td|jdd|jdfvr%td|jd|jtj|d z|z }t t ||}t|||\}}t|} tj|t } t|| | t|| |} | S) a[Tracks beats in an onset strength envelope. Parameters ---------- onset_envelope : np.ndarray [shape=(..., n,)] onset strength envelope bpm : float [scalar] or np.ndarray [shape=(...)] tempo estimate frame_rate : float [scalar] frame rate of the spectrogram (sr / hop_length, frames per second) tightness : float [scalar, positive] how closely do we adhere to bpm? trim : bool [scalar] trim leading/trailing beats with weak onsets? Returns ------- beats : np.ndarray [shape=(n,)] frame numbers of beat events rzbpm=z must be strictly positivez#tightness must be strictly positiver-rzInvalid bpm shape=z% does not match onset envelope shape=gN@r+) r6r8r r/round__beat_local_score__normalize_onsets__beat_track_dp __last_beatr=r>__dp_backtrack __trim_beats) rr"rfr r!frames_per_beat localscorebacklinkcumscoretailrIs rJrCrCs. vvcQhtC5(BCDDA~BCC yy}Q 4 4R 8991#))(t)F)nopythonrc ~t|}t|dk(rtjdtj|d |ddzdz|dz dzz}t|}t t|D]]}d||<t t d||dzz|z dzt ||dzz|D]!}||xx|||||dzz|z zz cc<#_yt|t|k(rt t|D]}tjdtj|| ||dzdz||z dzz}dt||zdz}d||<t t d||dzz|z dzt ||dzz|D]!}||xx|||||dzz|z zz cc<#yy)Nrgrg@@r*)lenr6exparangerBr\minr:)rrorpNwindowKiks rJriris NA ?q  ?1+=*=q?QTU?U VY] ]`opq`r rwxxxy Ks>*+AJqM3q!a1f*q.1"45s1qAv:q7IJ1 ^A1HaK-H!HH K , _ ^!4 4s>*+AVVDBIIq/A.A?STCUXYCY$Z]a$adstudv$v{|#||}FC*++a/AJqM3q!a1f*q.1"45s1qAv:q7IJ1 ^AQJN-K!KK K , 5rKz;void(float32[:], float32[:], float32, int32[:], float32[:])z;void(float64[:], float64[:], float32, int32[:], float64[:])z(t),(n),()->(t),(t)c 2d|jz}d}d|d<|d|d<tt|dkD}t|D]\}} tj } d} t |t j|||zdz z |d|||zzz dz dD]S} | dkrnL|| |t j|| z t j|||zz dzzz } | | kDsP| } | } U| dk\r | | z||<n| ||<|r | |krd||<| ||<d}y) z&Core dynamic program for beat trackingg{Gz?Tr-rrr{FN) r\r:r| enumerater6infrBrhlog)rpror rqrr score_thresh first_beattvrscore_i best_score beat_locationlocscores rJrkrk(s[*..**LJHQKQ-HQK S !A % &B + 7vvX  RXXob1f&=&ABBAO\^ab\bLcHcDcfgDgilmCQwSMIC266/Z\_`Z`JaCb1bef0f$ffEz!" # n A !J.HQK"HQK 'L0HQK'HQKJ7,rKz+void(float32[:], bool_[:], bool_, bool_[:])z+void(float64[:], bool_[:], bool_, bool_[:])z(t),(t),()->(t)c||ddtjd}tj|||t|dzt|t|dzz}|rd|dzj dzz}nd}d}|||krd||<|dz }|||krt|dz }|||krd||<|dz}|||kry) zCRemove spurious leading and trailing beats from the detection arrayNr{rTr*rFr)r6hanningconvolver|mean)rprIr! beats_trimmedw smooth_boe thresholdns rJrnrn[sM! 1 AZ.23q619S_SQRVUVY=VWJ JM//1367   A Q-9 $ a Q Q-9 $ J!A Q-9 $ a Q Q-9 $ rKcjtj|d}tjj ||}tjj |d}dtjj |z}tj|jddt}t|||||S)z/Identify the position of the last detected beatr-rW)datamaskrTNr.) rlocalmaxr6ma masked_arrayr7getdataemptyr/r:__last_beat_selector)rrr masked_scoresmedians thresholdsrss rJrlrls MM( , ,DEE&&H4&@Meell=rl2Gruu}}W--J 88(.."-S 9D4T: KrKz-void(float32[:], bool_[:], float32, int64[:])z-void(float64[:], bool_[:], float64, int64[:])z(t),(t),()->()crt|dz }||d<|dk\r||s|||k\r||d<y|dz}|dk\ryy)zmVectorized helper to identify the last valid beat position: cumscore[n] > threshold and not mask[n] rrN)r|)rrrroutrs rJrrsR H A CF q&Aw8A;)3CF  FA q&rKzvoid(int32[:], int32, bool_[:])zvoid(int64[:], int64, bool_[:])z (t),()->(t)c4|}|dk\rd||<||}|dk\ryy)z>Populate the beat indicator array from a sequence of backlinksrTN) backlinksrsrIrs rJrmrms+ A q&a aL q&rK)1__doc__numpyr6scipy scipy.statsnumba_cacherrrrfeaturerr r r?util.exceptionsr util.decoratorsr r typingrrrrr_typingr__all__ndarrayr<r:r>stats rv_continuousstrrrrCrj guvectorizerirkrnrlrrmrrKrJrsd  1$+-88" ( X-xQVW  #+/6:15] ] ]RZZ( ]  ]  ]] ] % rzz12 3] EKK-- .] ] ] 5 * +RZZ 78]D#+/!#!$15z* z* z*RZZ( z*  z*  z*z*z* EKK-- .z*ZZz*z3JJ3%'ZZ3=B3OT3\`3ZZ3n/  6 6  U $L$LD I I  T #)#)X55     <  ; ;  T       - -  T   rK