o jX2@sddlZddlZddlmZmZmZmZmZddlZddl Z ddl Z ddl m Z ddlmZddlmZdefddZd edefd d ZGd d d ZGdddeZdS)N)AnyDictListTupleUnion) load_config)setup_encoder_model)AudioProcessorpathcCs|dr"t|d }t|WdS1swYdS|drFt|d}tj|ddWdS1s?wYdStd)N.jsonr.pthrbcpu) map_locationUnsupported file type)endswithfsspecopenjsonloadtorch ValueError)r frI/home/kuhnn/.local/lib/python3.10/site-packages/TTS/tts/utils/managers.py load_files $  $robjcCs|dr&t|d}tj||ddWddS1swYdS|drJt|d}t||WddS1sCwYdStd)Nr windentr wbr)rrrrdumprsaver)rr rrrr save_files " "r%c@seZdZdZddefddZededefdd Zeded e dd fd d Z de dedd fddZ dedd fddZ dedd fddZdefddZede dedeefddZd S) BaseIDManagerz~Base `ID` Manager class. Every new `ID` manager must inherit this. It defines common `ID` manager specific functions. id_file_pathcCsi|_|r ||dSdSN) name_to_idload_ids_from_file)selfr(rrr__init__)szBaseIDManager.__init__json_file_pathreturncCs:t|d }t|WdS1swYdS)Nr )rrrr)r.rrrr _load_json/s$zBaseIDManager._load_jsondataNcCsBt|d}tj||ddWddS1swYdS)Nrrr )rrrr#)r.r1rrrr _save_json4s"zBaseIDManager._save_jsonitems parse_keycCs|j||d|_dS)z{Set IDs from data samples. Args: items (List): Data sampled returned by `load_tts_samples()`. r4N)parse_ids_from_datar*)r,r3r4rrrset_ids_from_data9szBaseIDManager.set_ids_from_data file_pathcCst||_dS)z[Set IDs from a file. Args: file_path (str): Path to the file. N)rr*r,r8rrrr+Asz BaseIDManager.load_ids_from_filecCt|j|dS)zfSave IDs to a json file. Args: file_path (str): Path to the output file. N)r%r*r9rrrsave_ids_to_fileIzBaseIDManager.save_ids_to_filecCs(|jr|jtt|jdSdS)dGet a random embedding. Args: Returns: np.ndarray: embedding. rN)r*randomchoiceslistkeysr,rrr get_random_idQszBaseIDManager.get_random_idcs,tfdd|D}ddt|D}|S)aParse IDs from data samples retured by `load_tts_samples()`. Args: items (list): Data sampled returned by `load_tts_samples()`. parse_key (str): The key to being used to parse the data. Returns: Tuple[Dict]: speaker IDs. csh|]}|qSrr).0itemr5rr hz4BaseIDManager.parse_ids_from_data..cSi|]\}}||qSrrrDinamerrr iz5BaseIDManager.parse_ids_from_data..)sorted enumerate)r3r4classesidsrr5rr6^s z!BaseIDManager.parse_ids_from_data)r')__name__ __module__ __qualname____doc__strr- staticmethodrr0dictr2rr7r+r;rrCrr6rrrrr&$s r&c seZdZdZ     d9deeeefdedededef fd d Ze d d Z e d dZ e ddZ e ddZ deddfddZedefddZdeddfddZdeeddfddZdedefd d!Zd"edeefd#d$Zdefd%d&Zd:d"ed'ed(edejfd)d*Zdefd+d,Zdefd-d.Zd;d/ed0eddfd1d2Zd3eeeefdefd4d5Z d6ee!j"ejfdefd7d8Z#Z$S)<EmbeddingManagerauBase `Embedding` Manager class. Every new `Embedding` manager must inherit this. It defines common `Embedding` manager specific functions. It expects embeddings files in the following format: :: { 'audio_file_key':{ 'name': 'category_name', 'embedding'[] }, ... } `audio_file_key` is a unique key to the audio file in the dataset. It can be the path to the file or any other unique key. `embedding` is the embedding vector of the audio file. `name` can be name of the speaker of the audio file. r'Fembedding_file_pathr(encoder_model_pathencoder_config_pathuse_cudacsxtj|di|_i|_g|_d|_d|_||_|r+t|t r&| |n| ||r8|r:| |||dSdSdS)N)r() superr- embeddingsembeddings_by_namesclip_idsencoder encoder_apr] isinstancer@"load_embeddings_from_list_of_filesload_embeddings_from_file init_encoder)r,rZr(r[r\r] __class__rrr-s   zEmbeddingManager.__init__cC t|jSzGet number of embeddings.)lenr_rBrrrnum_embeddings zEmbeddingManager.num_embeddingscCrjrk)rlr`rBrrr num_namesrnzEmbeddingManager.num_namescCs*|jrt|jt|jddSdS)zIDimensionality of embeddings. If embeddings are not loaded, returns zero.r embedding)r_rlr@rArBrrr embedding_dims zEmbeddingManager.embedding_dimcCt|jS)zGet embedding names.)r@r`rArBrrrembedding_namessz EmbeddingManager.embedding_namesr8r/NcCr:)zmSave embeddings to a json file. Args: file_path (str): Path to the output file. N)r%r_r9rrrsave_embeddings_to_filer<z(EmbeddingManager.save_embeddings_to_filecCst|}tdd|D}ddt|D}tttdd|D}i}|D]}|d|vrA|dg||d<q-||d|dq-||||fS) zhLoad embeddings from a json file. Args: file_path (str): Path to the file. cSsh|]}|dqS)rKr)rDxrrrrFrGz=EmbeddingManager.read_embeddings_from_file..cSrHrrrIrrrrLrMz>EmbeddingManager.read_embeddings_from_file..css|]}|VqdSr)r)rD clip_namerrr sz=EmbeddingManager.read_embeddings_from_file..rKrp)rrNvaluesrOr@setrAappend)r8r_speakersr*rar`rurrrread_embeddings_from_files  z*EmbeddingManager.read_embeddings_from_filecCs||\|_|_|_|_dS)ztLoad embeddings from a json file. Args: file_path (str): Path to the target json file. N)r|r*rar_r`r9rrrrfsz*EmbeddingManager.load_embeddings_from_file file_pathscCsi|_g|_i|_i|_|D]<}||\}}}}t|jt|@}|r2td|d||j||j ||j||j|qddt |jD|_dS)zLoad embeddings from a list of json files and don't allow duplicate keys. Args: file_paths (List[str]): List of paths to the target json files. z [!] Duplicate embedding names in cSrHrrrIrrrrLrMzGEmbeddingManager.load_embeddings_from_list_of_files..N) r*rar`r_r|ryrArupdateextendrO)r,r}r8rQrar_r` duplicatesrrrres   z3EmbeddingManager.load_embeddings_from_list_of_filesclip_idxcCs|j|dS)zGet embedding by clip ID. Args: clip_idx (str): Target clip ID. Returns: List: embedding as a list. rp)r_)r,rrrrget_embedding_by_clips z&EmbeddingManager.get_embedding_by_clipidxcCs |j|S)zGet all embeddings of a speaker. Args: idx (str): Target name. Returns: List[List]: all the embeddings of the given speaker. )r`)r,rrrrget_embeddings_by_names z'EmbeddingManager.get_embeddings_by_namecCsRi}|jD]}|d|vr|dg||d<q||d|dq|S)zmGet all embeddings by names. Returns: Dict: all the embeddings of each speaker. rKrp)r_rxrArz)r,r`rurrrget_embeddings_by_namess z(EmbeddingManager.get_embeddings_by_names num_samples randomizecCs||}|durt|d}|St||ks#Jd|d||r4ttj||dd}|St|d|d}|S)aSGet mean embedding of a idx. Args: idx (str): Target name. num_samples (int, optional): Number of samples to be averaged. Defaults to None. randomize (bool, optional): Pick random `num_samples` of embeddings. Defaults to False. Returns: np.ndarray: Mean embedding. Nrz [!] z has number of samples < )k)rnpstackmeanrlr>r?)r,rrrr_rrrget_mean_embeddings  z#EmbeddingManager.get_mean_embeddingcCs,|jr|jtt|jddSdS)r=rrpN)r_r>r?r@rArBrrrget_random_embedding+s"z%EmbeddingManager.get_random_embeddingcCrrr))rNr_rArBrrr get_clips8szEmbeddingManager.get_clips model_path config_pathcCsN||_t||_t|j|_|jj|j|d|dd|_tdi|jj|_ dS)zInitialize a speaker encoder model. Args: model_path (str): Model file path. config_path (str): Model config file path. use_cuda (bool, optional): Use CUDA. Defaults to False. T)evalr]cacheNr) r]rencoder_configrrbload_checkpointencoder_criterionr audiorc)r,rrr]rrrrg;s   zEmbeddingManager.init_encoderwav_filecspdtffdd }t|tr.d}|D]}||}|dur|}q||7}q|t|dS||}|dS)zCompute a embedding from a given audio file. Args: wav_file (Union[str, List[str]]): Target file path. Returns: list: Computed embedding. rcsnjj|jjd}jjddsj|}t|}nt|}j r*| }| d}j |}|S)N)sruse_torch_specFr)rcload_wav sample_rater model_paramsgetmelspectrogramr from_numpyr]cuda unsqueezerbcompute_embedding)rwaveformm_inputrprBrr_computeUs     z>EmbeddingManager.compute_embedding_from_clip.._computeNr)rVrdr@rltolist)r,rrr_wfrprrBrcompute_embedding_from_clipKs   z,EmbeddingManager.compute_embedding_from_clipfeatscCsDt|tjr t|}|jdkr|d}|jr|}|j |S)zCompute embedding from features. Args: feats (Union[torch.Tensor, np.ndarray]): Input features. Returns: List: computed embedding. r) rdrndarrayrrndimrr]rrbr)r,rrrrcompute_embeddingsps    z#EmbeddingManager.compute_embeddings)r'r'r'r'F)NF)F)%rRrSrTrUrrVrboolr-propertyrmrorqrsrtrWr|rfrerrrrintrrrrrrrgr@rrTensorr __classcell__rrrhrrYmsP        &%rY)rr>typingrrrrrrnumpyrr TTS.configrTTS.encoder.utils.generic_utilsrTTS.utils.audior rVrr%r&rYrrrrs     I