o 7wi@sdZddlZddlmZejfddZejddddfd d Zejfd d Zejfd dZddejfddZ ddejfddZ ejfddZ dS)zWER print functions. The functions here are used to print the computed statistics with human-readable formatting. They have a file argument, but you can also just use contextlib.redirect_stdout, which may give a nicer syntax. Authors * Aku Rouhe 2020 N) edit_distancecCsntdjd i||ddt|d|dkrdnd|dtdjd i||dtd jd i||dd S) arPrints out WER summary details in human-readable format. This function essentially mirrors the Kaldi compute-wer output format. Arguments --------- wer_details : dict Dict of wer summary details, see ``speechbrain.utils.edit_distance.wer_summary`` for format. file : stream Where to write. (default: sys.stdout) zl%WER {WER:.2f} [ {num_edits} / {num_scored_tokens}, {insertions} ins, {deletions} del, {substitutions} sub ])fileendnum_scored_sents num_ref_sentsz [PARTIAL]rz=%SER {SER:.2f} [ {num_erroneous_sents} / {num_scored_sents} ]zKScored {num_scored_sents} sentences, {num_absent_sents} not present in hyp.Nprintformat wer_detailsrr r S/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/speechbrain/dataio/wer.pyprint_wer_summarys2 rzz ; Tc Csh|r t|||d|D]&}|dr1|rt||dt|d|d|d|||d|r1t||dq dS)aPrint WER summary and alignments. Arguments --------- details_by_utterance : list List of wer details by utterance, see ``speechbrain.utils.edit_distance.wer_details_by_utterance`` for format. Has to have alignments included. file : stream Where to write. (default: sys.stdout) empty_symbol : str Symbol to use when aligning to nothing. separator : str String that separates each token in the output. Note the spaces in the default. print_header: bool Whether to print headers sample_separator: str A separator to put between samples (optional) r empty_symbol separatorscoredr alignment ref_tokens hyp_tokensN)_print_alignments_global_header_print_alignment_header_print_alignmentr )details_by_utterancerrr print_headersample_separatordetsr r rprint_alignments<s(  rcCstd|dtd|d|r&td|d|D]}tdjd i||dqntd|d|rGtd|d|D]}tdjd i||dq6dStd|ddS) NP================================================================================rzUTTERANCES WITH HIGHEST WERzANon-empty hypotheses -- utterances for which output was produced:z{key} %WER {WER:.2f}z(No utterances which had produced output!z@Empty hypotheses -- utterances for which no output was produced:z,No utterances which had not produced output!r r ) top_non_empty top_emptyrrr r r_print_top_wer_uttsns(   r#cCs>td|dtd|d|D]}tdjdi||dqdS)Nr rzSPEAKERS WITH HIGHEST WERz{speaker} %WER {WER:.2f}r r ) spks_by_werrrr r r_print_top_wer_spkss  r%cCsg}g}g}|D]E\} } } t| } | durt|| n|} | dur't|| n|}tt| t| t|}|| |||||| |qt|||dt|||dt|||ddS)Nr)strmaxlenappendcenterr join)rabrrra_paddedb_padded ops_paddedopij op_stringa_stringb_string pad_lengthr r rrsrcCstd|dtd|dtd|dtd|dtd|dgd}gd}tjd dd ftjd d d ftjd d dftjd ddftjd ddftjdddfg}t||||||ddS)Nr r ALIGNMENTSrzFormat:z, WER DETAILS) referenceonthefirstline)and hypothesisr:r;thirdinsrsubeqdelr)r r EDIT_SYMBOLSr)rrrr,r-rr r rrs,      rcCs(td|dtdjdi||ddS)Nr rzp{key}, %WER {WER:.2f} [ {num_edits} / {num_ref_tokens}, {insertions} ins, {deletions} del, {substitutions} sub ]r r r r r rrs  r) __doc__sysspeechbrain.utilsrstdoutrrr#r%rrrr r r rs" - 2