o 0i@@sPdZddlmZddlZddlZddlZddlmZddlm Z m Z ddl Z ddl ZddlmZmZmZmZmZmZmZmZmZmZmZmZmZmZmZddlm Z m!Z!m"Z"dd l#m$Z$e%e&Z'd8d9ddZ(d:ddZ)d;ddZ*d8dd?d"d#Z-d@d%d&Z.   dAdBd)d*Z/dCd+d,Z0dDd.d/Z1dEd2d3Z2dFd6d7Z3dS)GzModel loading and voice style management utilities. This module handles downloading, loading, and managing Supertonic TTS models and voice styles from HuggingFace Hub. ) annotationsN)Path)OptionalUnion) CFG_REL_PATHDEFAULT_CACHE_DIRDEFAULT_INTER_OP_NUM_THREADSDEFAULT_INTRA_OP_NUM_THREADSDEFAULT_MODEL_REPODEFAULT_MODEL_REVISIONDEFAULT_ONNX_PROVIDERSDP_ONNX_REL_PATHTEXT_ENC_ONNX_REL_PATHUNICODE_INDEXER_REL_PATHVECTOR_EST_ONNX_REL_PATHVOCODER_ONNX_REL_PATHVOICE_STYLES_DIRget_model_cache_dirget_model_repo)Style SupertonicUnicodeProcessor)validate_voice_style_format model_name Optional[str]returnrcCs<|dur t|}ntt}|jdddtd||S)aGet or create the cache directory for Supertonic models. Args: model_name: Model name ("supertonic" or "supertonic-2"). If None, uses default cache directory. Returns: Path object pointing to the cache directory Note: Default location is ~/.cache/supertonic2, but can be overridden with SUPERTONIC_CACHE_DIR environment variable NT)parentsexist_okzUsing cache directory: )rrrmkdirloggerdebug)r cache_dirr#K/home/kim/smarthome/.venv/lib/python3.10/site-packages/supertonic/loader.py get_cache_dir)s  r% list[Path]cCs ttttgS)zGet list of all required ONNX model file paths. Returns: List of Path objects for each required ONNX model file )rrrrr#r#r#r$"get_all_onnx_module_relative_paths@s r' model_dirUnion[Path, str]boolcs\ttr tnt}tfdd|D}|s,fdd|D}td||S)zCheck if all required ONNX model files exist in the directory. Args: model_dir: Directory to check for model files (str or Path) Returns: True if all required ONNX files exist, False otherwise c3s|] }|VqdSN)exists.0pathr(r#r$ Ysz'has_all_onnx_modules..cs g|] }|st|qSr#)r,strr-r0r#r$ \s z(has_all_onnx_modules..zMissing ONNX files: ) isinstancer2rr'allr r!)r(Zmodule_rel_pathsZ all_existmissingr#r0r$has_all_onnx_modulesNs r7Nonec Cszt|tr t|n|}|durt|}nt}|jd|jd}z=ddlm}t d|d|||t|t d| rLt d |t |t t|t|t d WdStys}z t d td |d}~wty}z>t d || rt dzt |Wnty}zt d|WYd}~nd}~wwtd|d||d}~ww)aDownload Supertonic model from HuggingFace Hub. Args: model_dir: Directory where the model should be downloaded (str or Path) model_name: Model name ("supertonic" or "supertonic-2"). If None, uses default model repository. N.z.tmpr)snapshot_downloadzDownloading model from z to temporary location: )repo_idZ local_dirrevisionz#Removing existing model directory: z%Model download completed successfullyzhuggingface_hub not installedzUFailed to import huggingface_hub. Please install it with: pip install huggingface-hubzModel download failed: zCleaning up temporary files...z$Failed to clean up temporary files: zFailed to download model from z>. Please check your internet connection and try again. Error: )r4r2rrr parentnameZhuggingface_hubr:r infor r,shutilrmtreemove ImportErrorerror RuntimeError Exceptionwarning)r(rr;temp_dirr:eZ cleanup_errorr#r#r$download_modelbsR     rJdictc Cst|tr t|n|}|t}|s#td|td|dz%t|d }t |}Wdn1s9wYt d||WSt j yf}ztd|t d|d |d}~ww) zLoad model configuration from JSON file. Args: model_dir: Directory containing the model files (str or Path) Returns: Dictionary containing model configuration zConfig file not found: z&Model configuration file not found at 1. Please ensure the model is properly downloaded.rNzLoaded config from zInvalid config format: z)Model configuration file is malformed at z. Please re-download the model.)r4r2rrr,r rDFileNotFoundErroropenjsonloadr!JSONDecodeError ValueError)r(Zcfg_pathfcfgsrIr#r#r$ load_configss,     rVintra_op_num_threads Optional[int]inter_op_num_threads]tuple[ort.InferenceSession, ort.InferenceSession, ort.InferenceSession, ort.InferenceSession]cs\dd d }t}tjj|_tjj|_|d ur|nt}|d ur!|nt }|d ur*||_ |d ur1||_ t }t fd d |D}|sOtd|ddg}ntd|d|d ur^|ndd|d urg|nd} td| |t} |t} |t} |t} td||| ||}|| ||}|| ||}|| ||}td||||fS)a0Load all ONNX model modules for TTS synthesis. Args: model_dir: Directory containing the ONNX model files (str or Path) intra_op_num_threads: Number of threads for intra-op parallelism. None (default) lets ONNX Runtime auto-detect optimal value inter_op_num_threads: Number of threads for inter-op parallelism. None (default) lets ONNX Runtime auto-detect optimal value Returns: Tuple of (duration_predictor, text_encoder, vector_estimator, vocoder) ONNX Runtime inference sessions onnx_pathroptsort.SessionOptions providers list[str]rort.InferenceSessioncSs8|s td|dtd|tj|||dS)zLoad a single ONNX model file.zONNX model file not found: rLzLoading ONNX model: )Z sess_optionsr^)r,rNr r!ortZInferenceSession)r[r\r^r#r#r$ _load_onnxs  z%load_onnx_modules.._load_onnxNcsg|]}|vr|qSr#r#)r.pZavailable_providersr#r$r3sz%load_onnx_modules..zRequested providers z4 not available. Falling back to CPUExecutionProviderZCPUExecutionProviderzUsing ONNX providers: zintra_threads=autoz, inter_threads=zONNX Runtime config: z$Loading ONNX models with providers: z#Successfully loaded all ONNX models)r[rr\r]r^r_rr`)raZSessionOptionsZGraphOptimizationLevelZORT_ENABLE_ALLZgraph_optimization_levelZ ExecutionModeZORT_SEQUENTIALZexecution_moder r rWrYr Zget_available_providersr rGr?rrrr)r(rWrYrbr\Z intra_threadsZ inter_threadsr^Zvalid_providers thread_infoZ dp_onnx_pathZtext_enc_onnx_pathZvector_est_onnx_pathZvocoder_onnx_pathdp_ort text_enc_ortvector_est_ort vocoder_ortr#rdr$load_onnx_modulessJ          rkrcCs>t|tr t|n|}|t}td|tt|}|S)zLoad the unicode text processor for the model. Args: model_dir: Directory containing the model files (str or Path) Returns: Initialized UnicodeProcessor instance zLoading text processor from )r4r2rrr r!r)r(Zunicode_indexer_pathtext_processorr#r#r$load_text_processors   rm auto_downloadrc Cstd|t|trt|n|}t|s.|s)td|td|dt||t |}t |||\}}}} t |} tdt || |||| S)aLoad the complete Supertonic TTS model. This function loads all model components including ONNX modules, configuration, and text processor. If model files are missing and auto_download is enabled, it will download them from HuggingFace Hub. Args: model_dir: Directory containing (or to contain) the model files (str or Path) auto_download: If True, automatically download missing model files intra_op_num_threads: Number of threads for intra-op parallelism. None (default) lets ONNX Runtime auto-detect optimal value inter_op_num_threads: Number of threads for inter-op parallelism. None (default) lets ONNX Runtime auto-detect optimal value model_name: Model name ("supertonic" or "supertonic-2"). Used for downloading the correct model if auto_download is True. Returns: Initialized Supertonic TTS engine zLoading model from zONNX models not found in zONNX model files not found in zz. Set auto_download=True to automatically download from HuggingFace Hub, or manually download the model to this directory.zModel loaded successfully) r r?r4r2rr7rDrNrJrVrkrmr) r(rnrWrYrrUrgrhrirjrlr#r#r$ load_model's     rocCsvt|tr t|n|}|t}|s#td|td|dtt | d}t dt |d||S)zList all available voice style JSON files in the model directory. Args: model_dir: Directory containing the model files (str or Path) Returns: Sorted list of paths to voice style JSON files z"Voice styles directory not found: z$Voice styles directory not found at rLz*.jsonzFound z voice styles in ) r4r2rrr,r rDrNsortedlistglobr!len)r(Zvoice_styles_dirpathsr#r#r$ list_available_voice_style_pathsXs  rur_cCst|}dd|D}|S)zList names of all available voice styles. Args: model_dir: Directory containing the model files (str or Path) Returns: List of voice style names (without .json extension) cSsg|]}|jqSr#)stemr-r#r#r$r3zsz4list_available_voice_style_names..)ru)r(Zvoice_style_pathsnamesr#r#r$ list_available_voice_style_namesps rxvoice_style_pathrc Cs&t|}ddd}|std|z@t|d }t|}Wd n1s*wYt|s;td |d t d |||d }||d}t ||WStj yr}zt d|td|d|d }~wt tfy}zt d|td|d||d }~ww)zLoad a voice style from a JSON file. Args: voice_style_path: Path to the voice style JSON file (str or Path) Returns: Style object with loaded style vectors json_datarKr np.ndarrayc SsRz|d}|d}tj|tjdj|WSty(}ztd||d}~ww)z"Parse style vector from JSON data.dimsdata)Zdtypez"Invalid style format: missing key N)nparrayZfloat32ZreshapeKeyErrorrS)rzr|r}rIr#r#r$_load_style_from_jsonsz>load_voice_style_from_json_file.._load_style_from_jsonzVoice style file not found: rMNzInvalid voice style format in zD. Expected 'style_ttl' and 'style_dp' with 'dims' and 'data' fields.zLoading voice style from Z style_ttlZstyle_dpz"Invalid JSON in voice style file: z!Voice style file is malformed at z. Please check the file format.zInvalid voice style format: z'Voice style file has invalid format at z5. Expected 'style_ttl' and 'style_dp' fields. Error: )rzrKrr{)rr,rNrOrPrQrrSr r!rrRrDr)ryrrTZvoice_style_jsonZ ttl_styleZdp_stylerIr#r#r$load_voice_style_from_json_file~sF         r voice_namer2cCslt|tr t|n|}|t|d}|s2t|}td|dtd|dd |t |S)a Load a voice style by name from the model directory. Args: model_dir: Directory containing the model files (str or Path) voice_name: Name of the voice style (without .json extension) Returns: Style object with loaded style vectors z.jsonz Voice style 'z ' not foundz%' not found. Available voice styles: z, ) r4r2rrr,rxr rDrNjoinr)r(rry availabler#r#r$load_voice_style_from_names rr+)rrrr)rr&)r(r)rr*)r(r)rrrr8)r(r)rrK)NN)r(r)rWrXrYrXrrZ)r(r)rr)NNN) r(r)rnr*rWrXrYrXrrrr)r(r)rr&)r(r)rr_)ryr)rr)r(r)rr2rr)4__doc__ __future__rrPloggingr@pathlibrtypingrrnumpyr~Z onnxruntimeraconfigrrr r r r r rrrrrrrrcorerrrutilsr getLogger__name__r r%r'r7rJrVrkrmrorurxrrr#r#r#r$s<  D       9" \ 1  7