hezar.preprocessors.audio_feature_extractor module

class hezar.preprocessors.audio_feature_extractor.AudioFeatureExtractor(config: AudioFeatureExtractorConfig, **kwargs)

Bases: Preprocessor

Base class for all audio feature extractors.

config_filename = 'feature_extractor_config.yaml'

classmethod load(hub_or_local_path, subfolder: str | None = None, config_filename: str | None = None, cache_dir: str | None = None, **kwargs)

Load a feature extractor from Hub or local path.

Parameters:

• hub_or_local_path – Hub repo id or local path

• subfolder – Preprocessor subfolder path

• config_filename – Config file name

• cache_dir – Path to cache directory

• **kwargs

Returns:

A AudioFeatureExtractor object

model_input_name = 'input_features'

pad(processed_features, padding: bool | str | PaddingType = True, max_length=None, truncation=None, pad_to_multiple_of=None, return_attention_mask=None, return_tensors=None)

Pad input values / input vectors or a batch of input values / input vectors up to predefined length or to the max sequence length in the batch.

Parameters:

• processed_features – Processed inputs to add padding to

• padding – Padding strategy which can be longest, max_length, False

• max_length – Max input length (Only effective if padding is max_length too, ignored otherwise)

• truncation – Whether to truncate long inputs or not

• pad_to_multiple_of – If set will pad the sequence to a multiple of the provided value.

• return_attention_mask – Whether to return the attention mask.

• return_tensors – Tensors return type among pt, np, list

push_to_hub(repo_id, subfolder=None, commit_message=None, private=None, config_filename=None)

Push the feature extractor files to a repo on the Hub.

Parameters:

• repo_id – Hub repo id

• subfolder – Subfolder to save, defaults to self.preprocessor_subfolder (preprocessor)

• commit_message – Commit message for the push

• private – If the repo does not exist already, specify whether the created repo must be private or not

• config_filename – Config filename, defaults to self.config_filename (feature_extractor_config.yaml)

save(path, subfolder=None, config_filename=None)

Save the feature extractor to the path. This normally is equal to only saving the feature_extractor_config.yaml file.

Parameters:

• path – Main path to save the feature extractor files

• subfolder – Optional subfolder, defaults to preprocessor

• config_filename – Optional config file name, defaults to feature_extractor_config.yaml

class hezar.preprocessors.audio_feature_extractor.AudioFeatureExtractorConfig(feature_size: 'int' = None, sampling_rate: 'int' = 16000, padding: 'str' = None, padding_value: 'float' = 0.0, padding_side: 'str' = None)

Bases: PreprocessorConfig

feature_size: int = None

padding: str = None

padding_side: str = None

padding_value: float = 0.0

sampling_rate: int = 16000