o ei@shdZddlZddlZddlmZmZmZmZddl m Z ddl m Z e eZGddde ZdgZdS) zXcodec model configurationN) AutoConfig DacConfig HubertConfig WavLMConfig)PreTrainedConfig)loggingcseZdZdZdZeedZdddddgddgddgdddd ddf d ee dBd e d e d ee dee dee de de de dBde de eBdBde e BdBffdd Z ede fddZede fddZede fddZede fdd Zede fd!d"Zede fd#d$ZZS)% XcodecConfiga This is the configuration class to store the configuration of an [`XcodecModel`]. It is used to instantiate a Xcodec model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the [Manel/X-Codec](https://huggingface.co/Manel/X-Codec) architecture. Configuration objects inherit from [`PreTrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PreTrainedConfig`] for more information. Args: target_bandwidths (`List[float]`, *optional*, defaults to `[0.5, 1, 1.5, 2, 4]`): The range of different bandwidths (in kbps) the model can encode audio with. sample_rate (`int`, *optional*, defaults to 16000): The sampling rate at which the audio waveform should be digitalized, in hertz (Hz). kernel_size (`int`, *optional*, defaults to 3): Kernel size for the initial semantic convolution. channel_ratios (`List[float]`, *optional*, defaults to `[1, 1]`): Expansion factors for the number of output channels in each semantic block. strides (`List[int]`, *optional*, defaults to `[1, 1]`): Strides for each semantic encoder block. block_dilations (`List[int]`, *optional*, defaults to `[1, 1]`): Dilation factors for the residual units in semantic blocks. unit_kernel_size (`int`, *optional*, defaults to 3): Kernel size inside each ResidualUnit in semantic blocks. codebook_size (`int`, *optional*, defaults to 1024): Number of entries in each residual quantizer's codebook. codebook_dim (`int`, *optional*): Dimensionality of each codebook vector. Defaults to sum of hidden size of acoustic and semantic models. initializer_range (`float`, *optional*, defaults to 0.02): Standard deviation of the truncated normal initializer for all weight matrices. acoustic_model_config (`Union[Dict, DacConfig]`, *optional*): An instance of the configuration for the acoustic (DAC) model. semantic_model_config (`Union[Dict, HubertConfig, WavLMConfig]`, *optional*): An instance of the configuration object for the semantic (HuBERT) model. Example: ```python >>> from transformers import XcodecModel, XcodecConfig >>> # Initializing configuration >>> configuration = XcodecConfig() >>> # Initializing a model (with random weights) from the configuration >>> model = XcodecModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```xcodec)acoustic_model_configsemantic_model_configNi>rg{Gz?target_bandwidths sample_rate kernel_sizechannel_ratiosstridesblock_dilationsunit_kernel_size codebook_size codebook_diminitializer_ranger r c  s`| durtdgddgddd|_n t| tr!td i| |_nt| tr*| |_n tdt| | dur transformersrrrrconfiguration_utilsrutilsr get_loggerrDr+r __all__r%r%r%r3s