h SSKJrJr SSKJrJr SSKrSSKJr SSKJ r SSKJ r "SS\5r "SS \ 5r SS \ S \\S \4S jjr"SS\5rSrSr"SS\5rg))ABCabstractmethod)DictListN)Tensor) TokenSpancB\rSrSr\S\\S\\\4Sj5rSrg) ITokenizer transcriptreturncg)zTokenize the given transcript (list of word) .. note:: The toranscript must be normalized. Args: transcript (list of str): Transcript (list of word). Returns: (list of int): List of token sequences N)selfr s ^/var/www/auris/envauris/lib/python3.13/site-packages/torchaudio/pipelines/_wav2vec2/aligner.py__call__ITokenizer.__call__ rN) __name__ __module__ __qualname____firstlineno__rrstrr__static_attributes__rrrr r s.  49  d3i    rr cP\rSrSrS\\\44SjrS\\S\\\4Sjr Sr g) Tokenizer dictionarycXlgNr)rrs r__init__Tokenizer.__init__s$rr r c zUVVs/sH!o"Vs/sHo0RUPM snPM# snn$s snfs snnfr!r")rr wordcs rrTokenizer.__call__s3?IJztT2T#T2zJJ2Js 72 77r"N) rrrrrrintr#rrrrrrrrs8%4S>%K49Kd3iKrremissiontokensblankcURnURS5n[R"U/[RUS9n[ R "XUS9upVUR5nUSUSpeXV4$)Nr)dtypedevicer,)r/ unsqueezetorchtensorint32F forced_alignexp)r*r+r,r/targetsaligned_tokensscoress r_align_emission_and_tokensr;#sj __F!!!$HllF85;;vFG^^HUKN ZZ\F+A.q F  !!rc L\rSrSr\S\S\\\S\\\4Sj5r Sr g)IAligner/r*r+r cg)aGenerate list of time-stamped token sequences Args: emission (Tensor): Sequence of token probability distributions in log-domain. Shape: `(time, tokens)`. tokens (list of integer sequence): Tokenized transcript. Output from :py:class:`torchaudio.pipelines.Wav2Vec2FABundle.Tokenizer`. Returns: (list of TokenSpan sequence): Tokens with time stamps and scores. Nr)rr*r+s rrIAligner.__call__0rrrN) rrrrrrrr)rrrrrrr=r=/s:    d3i  T$y/EZ    rr=c[U5[U5:XdeSn/nUHnURXX$-5 X$- nM U$)Nr)lensumappend)list_lengthsiretls r _unflattenrJ?sM u:W %% % A C  5QU#$  JrcHUVVs/sH oHo"PM M snn$s snnfr!r) nested_listrEitems r_flattenrNIs") rasq# !+  "K K " "c "3 "  s  = =h =r