h5jSSKrSSKrSSKJrJr SSKJr SSKJr SSK J r J r J r SSK r SSKrSSKJr SSKJrJrJr /rS\R,"\ R."\ R05R25-r\"S S \-5rS r"S S \ R<R>5r "SS\ R<R>5r!"SS\5r""SS\5r#"SS\ R<R>\"5r$"SS\#5r%\"SS55r&\&"S\"\SS9SSSS S!S"S#S$S%SS&9 r'S'\'l(g)(N)ABCabstractmethod) dataclass)partial)CallableListTuple) module_utils)emformer_rnnt_baseRNNTRNNTBeamSearch( g?c[R"X[R:5X[R:'X[R:*[R- X[R:*'U$N)torchlogmathexs Z/var/www/auris/envauris/lib/python3.13/site-packages/torchaudio/pipelines/rnnt_pipeline.py_piecewise_linear_logrsKIIaDFF m,A$&&jMDFF{^dff,A466kN Hc.^\rSrSrU4SjrSrSrU=r$)_FunctionalModulec.>[TU]5 Xlgr)super__init__ functional)selfr! __class__s rr _FunctionalModule.__init__s $rc$URU5$rr!r"inputs rforward_FunctionalModule.forwardsu%%rr&__name__ __module__ __qualname____firstlineno__r r)__static_attributes__ __classcell__r#s@rrrs%&&rrc.^\rSrSrU4SjrSrSrU=r$)_GlobalStatsNormalization#cX>[TU]5 [U5n[R"UR 55nSSS5 UR S[R"WS55 UR S[R"US55 g!,(df  Na=f)Nmean invstddev) rr openjsonloadsreadregister_bufferrtensor)r"global_stats_pathfblobr#s rr "_GlobalStatsNormalization.__init__$sx  # $::affh'D% VU\\$v,%?@ [%,,tK7H*IJ % $s %B B)c8XR- UR-$r)r7r8r's rr)!_GlobalStatsNormalization.forward-s !T^^33rr+r2s@rr4r4#sK44rr4cv\rSrSr\S\R S\\R \R 44Sj5rSr g)_FeatureExtractor1r(returncg)Generates features and length output from the given input tensor. Args: input (torch.Tensor): input tensor. Returns: (torch.Tensor, torch.Tensor): torch.Tensor: Features, with shape `(length, *)`. torch.Tensor: Length, with shape `(1,)`. NrEr's r__call___FeatureExtractor.__call__2rrEN) r,r-r.r/rrTensorr rLr0rErrrGrG1s8  ell  uU\\5<<5O/P    rrGc6\rSrSr\S\\S\4Sj5rSr g)_TokenProcessorBtokensrIc g)zDecodes given list of tokens to text sequence. Args: tokens (List[int]): list of tokens to decode. Returns: str: Decoded text sequence. NrE)r"rSkwargss rrL_TokenProcessor.__call__CrNrrEN) r,r-r.r/rrintstrrLr0rErrrQrQBs&  tCy  s    rrQc^\rSrSrSrS\R RSS4U4SjjrS\RS\ \R\R44Sjr S r U=r $) _ModuleFeatureExtractorPz``torch.nn.Module``-based feature extraction pipeline. Args: pipeline (torch.nn.Module): module that implements feature extraction logic. pipelinerINc.>[TU]5 Xlgr)rr r\)r"r\r#s rr _ModuleFeatureExtractor.__init__Ws  rr(crURU5n[R"URS/5nX#4$)rKr)r\rr>shape)r"r(featureslengths rr)_ModuleFeatureExtractor.forward[s5=='x~~a012r)r\)r,r-r.r/__doc__rnnModuler rOr r)r0r1r2s@rrZrZPsQ !!T! U\\ eELL%,,4N.O  rrZcJ\rSrSrSrS\SS4SjrS S\\S\ S\4S jjr S r g) _SentencePieceTokenProcessormzhSentencePiece-model-based token processor. Args: sp_model_path (str): path to SentencePiece model. sp_model_pathrINc$[R"S5(d [S5eSSKnUR US9UlUR R 5UR R5UR R51Ul g)N sentencepiecez2SentencePiece is not available. Please install it.r) model_file) r is_module_available RuntimeErrorrlSentencePieceProcessorsp_modelunk_ideos_idpad_idpost_process_remove_list)r"rjspms rr %_SentencePieceTokenProcessor.__init__tsq//@@ST T#22m2L MM " MM " MM ") %rrSlstripcUSSVs/sHo3UR;dMUPM nnSRURRU55R SS5nU(aUR 5$U$s snf)aDecodes given list of tokens to text sequence. Args: tokens (List[int]): list of tokens to decode. lstrip (bool, optional): if ``True``, returns text sequence with leading whitespace removed. (Default: ``True``). Returns: str: Decoded text sequence. Nu▁ )rujoinrq id_to_piecereplacerx)r"rSrx token_indexfiltered_hypo_tokens output_strings rrL%_SentencePieceTokenProcessor.__call__s},2!": +5KDLiLi9iK:   9 9:N OPXXYacfg  '') )  s A:A:)rurq)T) r,r-r.r/rdrXr rrWboolrLr0rErrrhrhms=  c  d  !tCy!$!#!!rrhc\rSrSr%Sr"SS\5r"SS\5r\ \ S'\ /\ 4\ S'\ \ S '\ \ S '\ \ S '\ \ S '\ \ S '\ \ S'\ \ S'\ \ S'\ \ S'\ \ S'S\ 4Sjr\S\ 4Sj5r\S\ 4Sj5r\S\ 4Sj5r\S\ 4Sj5r\S\ 4Sj5r\S\ 4Sj5rS\4SjrS\4SjrS\4SjrS\4SjrSrg )! RNNTBundleuDataclass that bundles components for performing automatic speech recognition (ASR, speech-to-text) inference with an RNN-T model. More specifically, the class provides methods that produce the featurization pipeline, decoder wrapping the specified RNN-T model, and output token post-processor that together constitute a complete end-to-end ASR inference pipeline that produces a text sequence given a raw waveform. It can support non-streaming (full-context) inference as well as streaming inference. Users should not directly instantiate objects of this class; rather, users should use the instances (representing pre-trained models) that exist within the module, e.g. :data:`torchaudio.pipelines.EMFORMER_RNNT_BASE_LIBRISPEECH`. Example >>> import torchaudio >>> from torchaudio.pipelines import EMFORMER_RNNT_BASE_LIBRISPEECH >>> import torch >>> >>> # Non-streaming inference. >>> # Build feature extractor, decoder with RNN-T model, and token processor. >>> feature_extractor = EMFORMER_RNNT_BASE_LIBRISPEECH.get_feature_extractor() 100%|███████████████████████████████| 3.81k/3.81k [00:00<00:00, 4.22MB/s] >>> decoder = EMFORMER_RNNT_BASE_LIBRISPEECH.get_decoder() Downloading: "https://download.pytorch.org/torchaudio/models/emformer_rnnt_base_librispeech.pt" 100%|███████████████████████████████| 293M/293M [00:07<00:00, 42.1MB/s] >>> token_processor = EMFORMER_RNNT_BASE_LIBRISPEECH.get_token_processor() 100%|███████████████████████████████| 295k/295k [00:00<00:00, 25.4MB/s] >>> >>> # Instantiate LibriSpeech dataset; retrieve waveform for first sample. >>> dataset = torchaudio.datasets.LIBRISPEECH("/home/librispeech", url="test-clean") >>> waveform = next(iter(dataset))[0].squeeze() >>> >>> with torch.no_grad(): >>> # Produce mel-scale spectrogram features. >>> features, length = feature_extractor(waveform) >>> >>> # Generate top-10 hypotheses. >>> hypotheses = decoder(features, length, 10) >>> >>> # For top hypothesis, convert predicted tokens to text. >>> text = token_processor(hypotheses[0][0]) >>> print(text) he hoped there would be stew for dinner turnips and carrots and bruised potatoes and fat mutton pieces to [...] >>> >>> >>> # Streaming inference. >>> hop_length = EMFORMER_RNNT_BASE_LIBRISPEECH.hop_length >>> num_samples_segment = EMFORMER_RNNT_BASE_LIBRISPEECH.segment_length * hop_length >>> num_samples_segment_right_context = ( >>> num_samples_segment + EMFORMER_RNNT_BASE_LIBRISPEECH.right_context_length * hop_length >>> ) >>> >>> # Build streaming inference feature extractor. >>> streaming_feature_extractor = EMFORMER_RNNT_BASE_LIBRISPEECH.get_streaming_feature_extractor() >>> >>> # Process same waveform as before, this time sequentially across overlapping segments >>> # to simulate streaming inference. Note the usage of ``streaming_feature_extractor`` and ``decoder.infer``. >>> state, hypothesis = None, None >>> for idx in range(0, len(waveform), num_samples_segment): >>> segment = waveform[idx: idx + num_samples_segment_right_context] >>> segment = torch.nn.functional.pad(segment, (0, num_samples_segment_right_context - len(segment))) >>> with torch.no_grad(): >>> features, length = streaming_feature_extractor(segment) >>> hypotheses, state = decoder.infer(features, length, 10, state=state, hypothesis=hypothesis) >>> hypothesis = hypotheses[0] >>> transcript = token_processor(hypothesis[0]) >>> if transcript: >>> print(transcript, end=" ", flush=True) he hoped there would be stew for dinner turn ips and car rots and bru 'd oes and fat mut ton pieces to [...] c\rSrSrSrSrg)RNNTBundle.FeatureExtractorz:Interface of the feature extraction part of RNN-T pipelinerENr,r-r.r/rdr0rErrFeatureExtractorrsHrrc\rSrSrSrSrg)RNNTBundle.TokenProcessorz7Interface of the token processor part of RNN-T pipelinerENrrErrTokenProcessorrsErr _rnnt_path_rnnt_factory_func_global_stats_path_sp_model_path_right_padding_blank _sample_rate_n_fft_n_mels _hop_length_segment_length_right_context_lengthrIcUR5n[RRUR5n[ R "U5nURU5 UR5 U$r) r torchaudioutilsdownload_assetrrloadload_state_dicteval)r"modelpath state_dicts r _get_modelRNNTBundle._get_modelsT'')..t?ZZ%  j)  rcUR$)zCSample rate (in cycles per second) of input waveforms. :type: int )rr"s r sample_rateRNNTBundle.sample_rates    rcUR$)z'Size of FFT window to use. :type: int )rrs rn_fftRNNTBundle.n_ffts {{rcUR$)zPNumber of mel spectrogram features to extract from input waveforms. :type: int )rrs rn_melsRNNTBundle.n_mels s ||rcUR$)zTNumber of samples between successive frames in input expected by model. :type: int )rrs r hop_lengthRNNTBundle.hop_lengths rcUR$)zDNumber of frames in segment in input expected by model. :type: int )rrs rsegment_lengthRNNTBundle.segment_lengths ###rcUR$)zSNumber of frames in right contextual block in input expected by model. :type: int )rrs rright_context_lengthRNNTBundle.right_context_length%s )))rcLUR5n[XR5$)z7Constructs RNN-T decoder. Returns: RNNTBeamSearch )rr r)r"rs r get_decoderRNNTBundle.get_decoder-s !e[[11rc ^[RRTR5n[ [ R R[RRTRTRTRTRS9[S5[S5[U5[U4Sj555$)zbConstructs feature extractor for non-streaming (full-context) ASR. Returns: FeatureExtractor rrrrc&URSS5$Nrzr transposers r2RNNTBundle.get_feature_extractor..BAKK1,=rc&[U[-5$rr_gainrs rrrC,A!e),Lrct>[RRRUSSSTR45$)Nr)rrer!padr)rr"s rrrEs,EHH,?,?,C,CA1aQUQdQdGe,frrrrrrZrre Sequential transformsMelSpectrogramrrrrrr4r" local_paths` rget_feature_extractor RNNTBundle.get_feature_extractor6s  %%44T5L5LM & HH  %%44 $ 0 0 4;;cgcrcr5""=>!"LM)*5!"fg   rc p[RRUR5n[ [ R R[RRURURURURS9[S5[S5[U555$)z^Constructs feature extractor for streaming (simultaneous) ASR. Returns: FeatureExtractor rc&URSS5$rrrs rr.Urrc&[U[-5$rrrs rrrVrrrrs rget_streaming_feature_extractor*RNNTBundle.get_streaming_feature_extractorIs  %%44T5L5LM & HH  %%44 $ 0 0 4;;cgcrcr5""=>!"LM)*5    rcj[RRUR5n[ U5$)z9Constructs token processor. Returns: TokenProcessor )rrrrrhrs rget_token_processorRNNTBundle.get_token_processor[s+  %%44T5H5HI +J77rrEN)r,r-r.r/rdrGrrQrrX__annotations__rr rWrpropertyrrrrrrr rrrrr0rErrrrsUFPI,IFFO T** K K LD!S!!s C  $$$*c**2^2 '7 & 1A $8^8rrz(models/emformer_rnnt_base_librispeech.pti) num_symbolsz2pipeline-assets/global_stats_rnnt_librispeech.jsonz.pipeline-assets/spm_bpe_4096_librispeech.modelii>ir[) rrrrrrrrrrrraASR pipeline based on Emformer-RNNT, pretrained on *LibriSpeech* dataset :cite:`7178964`, capable of performing both streaming and non-streaming inference. The underlying model is constructed by :py:func:`torchaudio.models.emformer_rnnt_base` and utilizes weights trained on LibriSpeech using training script ``train.py`` `here `__ with default arguments. Please refer to :py:class:`RNNTBundle` for usage instructions. ))r:rabcrr dataclassesr functoolsrtypingrrr rrtorchaudio._internalr torchaudio.modelsr r r __all__log10iinfoint16max_decibelpowrrrerfrr4rGrQrZrhrEMFORMER_RNNT_BASE_LIBRISPEECHrdrErrrs0 #!(( -FF  DJJu{{5;;7;;< < Bx  && 4 4  "  c   ehhoo/@ :(!?(!V I8I8 I8X",91tDKC    " *&r