:j$bddlZddlmZmZddlmZmZmZmZddl Z ddl Z ddl Z ddlmcmZddlmZddlmZeddeedzdefd Zd Zd Zej4 dd e j6d e j6deedzdedee j6eeeff dZej4 dd e j<d e j<deedzdedee j<eeeff dZdefdede dee j6e j6ge j6fde jBfdZ"y)N)partialsingledispatch)CallableListLiteralTuple)SlidingWindowFeature)linear_sum_assignment cost_func)msemae return_costct)aFind cost-minimizing permutation Parameters ---------- y1 : np.ndarray or torch.Tensor (batch_size, num_samples, num_classes_1) y2 : np.ndarray or torch.Tensor (num_samples, num_classes_2) or (batch_size, num_samples, num_classes_2) cost_func : callable or {"mse", "mae"}, optional Can be either "mse" (mean squared error) or "mae" (mean absolute error) or a callable. When callable, takes two (num_samples, num_classes) sequences and returns (num_classes, ) pairwise cost. Defaults to computing mean squared error ("mse"). return_cost : bool, optional Whether to return cost matrix. Defaults to False. Returns ------- permutated_y2 : np.ndarray or torch.Tensor (batch_size, num_samples, num_classes_1) permutations : list of tuple List of permutations so that permutation[i] == j indicates that jth speaker of y2 should be mapped to ith speaker of y1. permutation[i] == None when none of y2 speakers is mapped to ith speaker of y1. cost : np.ndarray or torch.Tensor, optional (batch_size, num_classes_1, num_classes_2) ) TypeError)y1y2r rs u/home/mcse/projects/srt_converter/srt-converter-venv/lib/python3.12/site-packages/pyannote/audio/utils/permutation.py permutater%s : +c \tjtj||ddS)zCompute class-wise mean-squared error Parameters ---------- Y, y : (num_frames, num_classes) torch.tensor Returns ------- mse : (num_classes, ) torch.tensor Mean-squared error none) reductionraxis)torchmeanFmse_lossYykwargss r mse_cost_funcr#Es" ::ajjA8q AArc \tjtj||z dS)zCompute class-wise mean absolute difference error Parameters ---------- Y, y: (num_frames, num_classes) torch.tensor Returns ------- mae : (num_classes, ) torch.tensor Mean absolute difference error rr)rrabsrs r mae_cost_funcr&Ts" ::eiiA&Q //rrrreturnc|j\}}}t|jdk(r|j|dd}t|jdk7r d}t||j\}} } ||k7s|| k7r:dt |jdt |jd}t||d}g} g} |rg} t j |j|j|j } tt||D]\}\}}t j5|dk(r>|jd|jd z }t j||zd }n|d k(rN|jd|jd z }t jt j|d }nOt jt!|Dcgc]'}|||dd||d zfjd| )c}}ddd| |kDr6t#j$d d d | |z fdt j&|d z}n}dg|z}tt)|j+D]!\}}||ks |||<|dd|f| |dd|f<#| j-t ||s j-||r| | t j fS| | fScc}w#1swYxYw)NzAIncorrect shape: should be (batch_size, num_frames, num_classes).zShape mismatch: z vs. .r )devicedtyper)dimr constant)shapelenexpand ValueErrortuplerzerosr-r. enumeratezipno_grad unsqueezerr%stackrangerpadmaxr cpuappend)rrr r batch_size num_samples num_classes_1msg batch_size_ num_samples_ num_classes_2 permutations permutated_y2costsby1_y2_diffcosti padded_cost permutationk1k2s rpermutate_torchrVcs.0XX*J ] 288} YYz2r * 288}Qo/1xx,K}[ K<$? rxx 1uRXX6GqIo LMKK"((KM"3r2;/$ :C]]_ E!}}Q'#--*::zz$+15e#}}Q'#--*::zz%))D/q9{{"'}!5"#s1a!a%i<'8'?'?M'RS  = (%%Aq--78 $!# KKf}, 01BCD 5FBM!"$ B*-ae* aBh' 5 E+./  LL I$LlEKK,>>> , &&;  s*B3K3,K. K3.K33K< cttj|tj|||}|r'|\}}}|j||jfS|\}}|j|fS)Nr r)rr from_numpynumpy)rrr routputrJrIrKs rpermutate_numpyr\s{   F-3* |U""$lEKKMAA"(M<    , ..rg? segmentationsonsetc t||}|j}|jj\}}}t j |j |jz dz }d|fz}tj} t|D]a\} \} } ttd| |dz t|| |dzdzD]%} | | k(r t| | z |z|jz|j z }|dkr| }| |d}|| d||z f}n| d||z }|| |df}t|t j"||d\}\}\}t|D]\}}t!j$|dd|f|kD}t!j$|dd|f|kD}|r| j'| |f|r| j'| |f|sq|st| j)| |f| |f|||f(d| S) aaBuild permutation graph Parameters ---------- segmentations : (num_chunks, num_frames, local_num_speakers)-shaped SlidingWindowFeature Raw output of segmentation model. onset : float, optionan Threshold above which a speaker is considered active. Defaults to 0.5 cost_func : callable Cost function used to find the optimal bijective mapping between speaker activations of two overlapping chunks. Expects two (num_frames, num_classes) torch.tensor as input and returns cost as a (num_classes, ) torch.tensor. Defaults to mae_cost_func. Returns ------- permutation_graph : nx.Graph Nodes are (chunk_idx, speaker_idx) tuples. An edge between two nodes indicate that those are likely to be the same speaker (the lower the value of "cost" attribute, the more likely). )r^r/r)rNTrX)rP)rsliding_windowdatar2mathfloordurationstepnxGraphr8r=r?minroundrnpnewaxisanyadd_nodeadd_edge)r]r^r chunks num_chunks num_frames_ max_lookahead lookaheadpermutation_graphCchunk segmentationcshiftthis_segmentationsthat_segmentationsrSrPthisthatthis_is_activethat_is_actives rbuild_permutation_graphrs44 /I  ) )F - 2 2 8 8J AJJvrs2 -11 .0GL,A!AX\> B 0 38 I' I' I'',//I' I'  5<<eCj) )* I'I'X 38 / / /',/// /  2::tE#J' '( //0FSL'L Lu||4ellBCLXX Lr