o %ݫij @sdZddlZddlZddlmZddlZddlZddlmZddl m Z ddl m Z ddl mZddlmZmZdd lmZdd lmZdd lmZdd lmZdd lmZmZddlmZddl m!Z!ee"Z#dddidddddej$f defddZ%Gdddej&j'Z(GdddZ)dS)aKDefines interfaces for simple inference with pretrained models Authors: * Aku Rouhe 2021 * Peter Plantinga 2021 * Loren Lugosch 2020 * Mirco Ravanelli 2020 * Titouan Parcollet 2021 * Abdel Heba 2021 * Andreas Nautsch 2022, 2023 * Pooneh Mousavi 2023 * Sylvain de Langen 2023 * Adel Moumen 2023 * Pradnya Kandarkar 2023 N)SimpleNamespace)load_hyperpyyaml) DataParallel) SyncBatchNorm)DistributedDataParallel) PaddedBatch PaddedData)AudioNormalizer) DataPipeline) split_path) run_on_main) LocalStrategyfetch) get_logger)import_from_pathhyperparams.yaml custom.pyCustomInterfaceTFlocal_strategyc  Kst|||dd|d| | d } t|||dd|d| | d } tjt| jt| dd}t|||}Wdn1s:wY||d<|d}||t |j ||| dd |sr| t | }t ||}|d |d |d | SdS) a Fetch and load an interface from an outside source The source can be a location on the filesystem or online/huggingface The pymodule file should contain a class with the given classname. An instance of that class is returned. The idea is to have a custom Pretrained subclass in the file. The pymodule file is also added to the python path before the Hyperparams YAML file is loaded, so it can contain any custom implementations that are needed. The hyperparams file should contain a "modules" key, which is a dictionary of torch modules used for computation. The hyperparams file should contain a "pretrainer" key, which is a speechbrain.utils.parameter_transfer.Pretrainer Arguments --------- source : str or Path or FetchSource The location to use for finding the model. See ``speechbrain.utils.fetching.fetch`` for details. hparams_file : str The name of the hyperparameters file to use for constructing the modules necessary for inference. Must contain two keys: "modules" and "pretrainer", as described. pymodule_file : str The name of the Python file that should be fetched. classname : str The name of the Class, of which an instance is created and returned overrides : dict Any changes to make to the hparams file when it is loaded. overrides_must_match : bool Whether an error will be thrown when an override does not match a corresponding key in the yaml_stream. savedir : str or Path Where to put the pretraining material. If not given, just use cache. use_auth_token : bool (default: False) If true Huggingface's auth_token will be used to load private models from the HuggingFace Hub, default is False because the majority of models are public. download_only : bool (default: False) If true, class and instance creation is skipped. huggingface_cache_dir : str Path to HuggingFace cache; if None -> "~/.cache/huggingface" (default: None) local_strategy : speechbrain.utils.fetching.LocalStrategy The fetching strategy to use, which controls the behavior of remote file fetching with regards to symlinking and copying. See :func:`speechbrain.utils.fetching.fetch` for further details. **kwargs : dict Arguments to forward to class constructor. Returns ------- object An instance of a class with the given classname from the given pymodule file. FN filenamesourcesavedir overwrite save_filenameuse_auth_tokenrevisionhuggingface_cache_dirrutf-8encodingr pretrainerdefault_sourcerrkwargsmodules)r&hparams)rsyspathappendstrparentopenrset_collect_inr collect_filesload_collectedrgetattr)r hparams_file pymodule_file classname overridesoverrides_must_matchrr download_onlyrrr%hparams_local_pathpymodule_local_pathfinr'r!moduleclsr(r(T/home/ubuntu/.local/lib/python3.10/site-packages/speechbrain/inference/interfaces.py foreign_class(sTE     r?c seZdZdZgZgZ dfdd ZddZddd Zd d Z d d Z ddZ e ddidddddde jf de fddZZS) PretrainedaTakes a trained model and makes predictions on new data. This is a base class which handles some common boilerplate. It intentionally has an interface similar to ``Brain`` - these base classes handle similar things. Subclasses of Pretrained should implement the actual logic of how the pretrained system runs, and add methods with descriptive names (e.g. transcribe_file() for ASR). Pretrained is a torch.nn.Module so that methods like .to() or .eval() can work. Subclasses should provide a suitable forward() implementation: by convention, it should be a method that takes a batch of audio signals and runs the full model (as applicable). Arguments --------- modules : dict of str:torch.nn.Module pairs The Torch modules that make up the learned system. These can be treated in special ways (put on the right device, frozen, etc.). These are available as attributes under ``self.mods``, like self.mods.model(x) hparams : dict Each key:value pair should consist of a string key and a hyperparameter that is used within the overridden methods. These will be accessible via an ``hparams`` attribute, using "dot" notation: e.g., self.hparams.model(x). run_opts : dict Options parsed from command line. See ``speechbrain.parse_arguments()``. List that are supported here: * device * data_parallel_count * data_parallel_backend * distributed_launch * distributed_backend * jit * jit_module_keys * compule * compile_module_keys * compile_mode * compile_using_fullgraph * compile_using_dynamic_shape_tracing freeze_params : bool To freeze (requires_grad=False) parameters or not. Normally in inference you want to freeze the params. Also calls .eval() on all modules. NTc s,tddddddddddddd }|D],\}}|dur-||vr-t||||q|dur>||vr>t||||qt|||qtj||_|jD] }|dur]| |j qQ|j ri|durit d|dur|j D]} | |vr~t d| d qpt d i||_|||d t|_dS) NcpuFncclzreduce-overhead) devicedata_parallel_countdata_parallel_backenddistributed_launchdistributed_backendjitjit_module_keyscompilecompile_module_keys compile_modecompile_using_fullgraph#compile_using_dynamic_shape_tracingzNeed to provide hparams dict.zNeed hparams['z']audio_normalizerr()super__init__itemssetattrtorchnn ModuleDictmodsvaluestorDHPARAMS_NEEDED ValueErrorrr'_prepare_modulesgetr rP) selfr&r'run_opts freeze_paramsrun_opt_defaultsargdefaultr<hp __class__r(r>rRsJ     zPretrained.__init__cCs<|||r|j|jD]}d|_qdSdS)zPrepare modules for computation, e.g. jit. Arguments --------- freeze_params : bool Whether to freeze the parameters and call ``eval()``. FN)_compile_wrap_distributedrXeval parameters requires_grad)r_rapr(r(r>r] s  zPretrained._prepare_modulescCsLt|\}}t|||tjd}tjt|dd\}}||j}| ||S)aLoad an audio file with this model's input spec When using a speech model, it is important to use the same type of data, as was used to train the model. This means for example using the same sampling rate and number of channels. It is, however, possible to convert a file from a higher sampling rate to a lower one (downsampling). Similarly, it is simple to downmix a stereo file to mono. The path can be a local path, a web url, or a link to a huggingface repo. )rrrF)channels_first) r rr SYMLINK torchaudioloadr,rZrDrP)r_r*rrflsignalsrr(r(r> load_audios   zPretrained.load_audioc Csjttd}|s|jdurtdt}|jr+|jdur!t|j}n t|j}tdt}|j rF|j durrh3sd             zPretrained._compilecCstd|dS)Nz4'_compile_jit' is deprecated; use '_compile' instead)warningswarnrhr_r(r(r> _compile_jitts  zPretrained._compile_jitcCs|js|jsdS|jr4|jD]!\}}tdd|Dr1t|}t||j gd}||j|<qdS|jD]+\}}tdd|Drd|j dkrRt |}n t |ddt |j D}||j|<q9dS) z5Wrap modules with distributed wrapper when requested.Ncs|]}|jVqdSNrl.0rmr(r(r> ~z/Pretrained._wrap_distributed..) device_idscsrrrrr(r(r>rrrBcSsg|]}|qSr(r()rir(r(r> sz0Pretrained._wrap_distributed..) rGrFrXrSanyrkrconvert_sync_batchnormDDPrDrEDPrange)r_rr<r(r(r>rixs(      zPretrained._wrap_distributedrrFrc  Kst|||dd||| | d } zt|||dd||| | d }tjt|jWnty4|dkr1nYnwt| dd}t||| d}Wdn1sMwY||d<| d d}|dur| |t |j ||| d d |s| ||d |fi| SdS||d |fi| S) af Fetch and load based from outside source based on HyperPyYAML file The source can be a location on the filesystem or online/huggingface You can use the pymodule_file to include any custom implementations that are needed: if that file exists, then its location is added to sys.path before Hyperparams YAML is loaded, so it can be referenced in the YAML. The hyperparams file should contain a "modules" key, which is a dictionary of torch modules used for computation. The hyperparams file should contain a "pretrainer" key, which is a speechbrain.utils.parameter_transfer.Pretrainer Arguments --------- source : str The location to use for finding the model. See ``speechbrain.utils.fetching.fetch`` for details. hparams_file : str The name of the hyperparameters file to use for constructing the modules necessary for inference. Must contain two keys: "modules" and "pretrainer", as described. pymodule_file : str A Python file can be fetched. This allows any custom implementations to be included. The file's location is added to sys.path before the hyperparams YAML file is loaded, so it can be referenced in YAML. This is optional, but has a default: "custom.py". If the default file is not found, this is simply ignored, but if you give a different filename, then this will raise in case the file is not found. overrides : dict Any changes to make to the hparams file when it is loaded. savedir : str or Path Where to put the pretraining material. If not given, just use cache. use_auth_token : bool (default: False) If true Huggingface's auth_token will be used to load private models from the HuggingFace Hub, default is False because the majority of models are public. revision : str The model revision corresponding to the HuggingFace Hub model revision. This is particularly useful if you wish to pin your code to a particular version of a model hosted at HuggingFace. download_only : bool (default: False) If true, class and instance creation is skipped. huggingface_cache_dir : str Path to HuggingFace cache; if None -> "~/.cache/huggingface" (default: None) overrides_must_match : bool Whether the overrides must match the parameters already in the file. local_strategy : LocalStrategy, optional Which strategy to use to deal with files locally. (default: `LocalStrategy.SYMLINK`) **kwargs : dict Arguments to forward to class constructor. Returns ------- Instance of cls FNrrrr)r7rr!r"r$r&)rr)r*r+r,r-r\r.rr^r/r r0r1)r=rr3r4r6rrrr8rr7rr%r9r:r;r'r!r(r(r> from_hparamssfL      zPretrained.from_hparams)NNNTr)__name__ __module__ __qualname____doc__r[MODULES_NEEDEDrRr]rurhrri classmethodr ror __classcell__r(r(rfr>r@s2.8 A r@c@seZdZdZddZddZddZdd Zd d Zd d Z ddZ ddZ e ddZ e ddZe ddZddZddZddZdS)EncodeDecodePipelineMixinzv A mixin for pretrained models that makes it possible to specify an encoding pipeline and a decoding pipeline cCs`||jj||jjt|j|jjd|jjdd|_t|jj|jjd|jd|_dS)z< Initializes the encode and decode pipeline steps output_keys)static_data_keys dynamic_itemsrN)_run_init_stepsr'encode_pipelinedecode_pipeliner INPUT_STATIC_KEYSmodel_output_keys OUTPUT_KEYSrr(r(r>create_pipelines%s    z*EncodeDecodePipelineMixin.create_pipelinescCs>|dg}|D]}|d}|rt|std|qdS)zEncode/decode pipelines may include initialization steps, such as filling text encoders with tokens. Calling this method will run them, if definedinitfuncz Invalid pipeline init definitionN)r^callabler\)r_pipeline_definitionrstep step_funcr(r(r>r6s   z)EncodeDecodePipelineMixin._run_init_stepscs&|r|}|Sfdd|D}|S)Ncsg|]}|qSr(r()ritempipeliner(r>rEz;EncodeDecodePipelineMixin._run_pipeline..r()r_rinputbatchoutputr(rr> _run_pipelineAs z'EncodeDecodePipelineMixin._run_pipelinecCs|jr|S||Sr) batch_inputs_itemize)r_rr(r(r>_get_encode_pipeline_inputHsz4EncodeDecodePipelineMixin._get_encode_pipeline_inputcCsJt|jdd}|}t|dkr|f}|rtt||}|js#||}|S)Nr)r2r'lendictzip batch_outputsr)r_ model_outputrpipeline_inputr(r(r>_get_decode_pipeline_inputKs  z4EncodeDecodePipelineMixin._get_decode_pipeline_inputcsBtt|}||t|}fddt|DS)Ncs(g|]ttfddDqS)csg|]}|qSr(r()rvalueidxr(r>rarzAEncodeDecodePipelineMixin._itemize...)rr)rkeysrYrr>r`sz6EncodeDecodePipelineMixin._itemize..)nextiterrYrrr)r_r first_item batch_lengthr(rr>r\s  z"EncodeDecodePipelineMixin._itemizecs*ttrfddjjdDS)a Converts padded batches to dictionaries, leaves other data types as is Arguments --------- data: object a dictionary or a padded batch Returns ------- results: dict the dictionary csi|] }||qSr() _get_value)rkeydatar_r(r> us z5EncodeDecodePipelineMixin.to_dict..r) isinstancerr'r)r_rr(rr>to_dictes   z!EncodeDecodePipelineMixin.to_dictcCs$t||}|jst|tr|j}|S)aC Retrieves the value associated with the specified key, dereferencing .data where applicable Arguments --------- data: PaddedBatch a padded batch key: str the key Returns ------- result: object the result )r2input_use_padded_datarrr)r_rrrr(r(r>r{s z$EncodeDecodePipelineMixin._get_valuecC|jjddS)z Determines whether the input pipeline operates on batches or individual examples (true means batched) Returns ------- batch_inputs: bool rTr'rr^rr(r(r>r z&EncodeDecodePipelineMixin.batch_inputscCr)z If turned on, raw PaddedData instances will be passed to the model. If turned off, only .data will be used Returns ------- result: bool whether padded data is used as is use_padded_dataFrrr(r(r>rrz/EncodeDecodePipelineMixin.input_use_padded_datacCr)z Determines whether the output pipeline operates on batches or individual examples (true means batched) Returns ------- batch_outputs: bool rT)r'rr^rr(r(r>rrz'EncodeDecodePipelineMixin.batch_outputscCs |jst|jdt}||}|S)N collate_fn)rr2r'r)r_rrr(r(r>_collatesz"EncodeDecodePipelineMixin._collatecCsH||}|j|j||jd}||}t|dr||j}||S)z Encodes the inputs using the pipeline Arguments --------- input: dict the raw inputs Returns ------- results: object rrrrZ) rrrrrrzrZrDr)r_rr model_inputr(r(r> encode_inputs     z&EncodeDecodePipelineMixin.encode_inputcCs||}|j|j||jdS)z Decodes the raw model outputs Arguments --------- output: tuple raw model outputs Returns ------- result: dict or list the output of the pipeline r)rrrr)r_rrr(r(r> decode_outputs z'EncodeDecodePipelineMixin.decode_outputN)rrrrrrrrrrrrpropertyrrrrrrr(r(r(r>r s&       r)*rr)rtypesrrUrp hyperpyyamlrtorch.nnrrrtorch.nn.parallelrrspeechbrain.dataio.batchrrspeechbrain.dataio.preprocessr speechbrain.utils.data_pipeliner speechbrain.utils.data_utilsr speechbrain.utils.distributedr speechbrain.utils.fetchingr rspeechbrain.utils.loggerrspeechbrain.utils.superpowersrrr|ror?rVModuler@rr(r(r(r>sJ            y