o Si@sddlmZmZddlmZmZddlZddlm Z m Z ddl m Z m Z mZmZmZeGdddZde fd ejd ed ed efd dZddddddde fdejdededededededededefddZe Gd d!d!e ZdS)")asdict dataclass)AnyDictN)FeatureExtractorregister_extractor)EPSILON LOG_EPSILONSecondscompute_num_framesis_module_availablec@seZdZUdZdZeed<dZeed<dZeed<dZ eed <d Z e ed <d Z eed <d Z eed<dZeed<dee effddZedee efddfddZdS)LibrosaFbankConfiga Default librosa config with values consistent with various TTS projects. This config is intended for use with popular TTS projects such as [ParallelWaveGAN](https://github.com/kan-bayashi/ParallelWaveGAN) Warning: You may need to normalize your features. i"V sampling_ratefft_sizehop_sizeN win_lengthhannwindowP num_mel_binsfminfmaxreturncCst|SN)rselfrQ/home/ubuntu/.local/lib/python3.10/site-packages/lhotse/features/librosa_fbank.pyto_dict!zLibrosaFbankConfig.to_dictdatacCstdi|S)Nr)r )r#rrr from_dict$szLibrosaFbankConfig.from_dict)__name__ __module__ __qualname____doc__rint__annotations__rrrrstrrrrrrr! staticmethodr$rrrr r s          r featsexpected_num_framesabs_tol pad_valuecCs|jd|}d|kr|krnn|d|}|S| |kr&dkr8nntj|d| fdfdtd}|St||krKtd|d|jd|S)Nr)rrconstant)modeconstant_valuesz Expected z source_feats; feats.shape[0] = )shapenppadr abs ValueError)r.r/r0r1 frames_diffrrr pad_or_truncate_features)s"    r;rrrrraudiorrrrrrrrepsc  Cstdr ddl} ntdt|jdkr)|jddks$Jd|jd|d}nt|jdks9Jd|jd| j|||||d d } t| j} |durPdn|}|durZ|dn|}| j j |||||d } t t | t | | j}tt|||||d }t||}|S) aCompute log-Mel filterbank feature. Args: audio (ndarray): Audio signal (T,). sampling_rate (int): Sampling rate. fft_size (int): FFT size. hop_size (int): Hop size. win_length (int): Window length. If set to None, it will be the same as fft_size. window (str): Window function type. num_mel_bins (int): Number of mel basis. fmin (int): Minimum frequency in mel basis calculation. fmax (int): Maximum frequency in mel basis calculation. eps (float): Epsilon value to avoid inf in log calculation. Returns: ndarray: Log Mel filterbank feature (#source_feats, num_mel_bins). librosarNzULibrosa is not installed. Please install librosa before using LibrosaFbank extractor.r-z?LibrosaFbank works only with single-channel recordings (shape: )reflect)n_fft hop_lengthrrpad_mode)srrBn_melsrr)duration frame_shiftr)r r> ImportErrorlenr5stftr6r8Tfiltersmellog10maximumdotr r;)r<rrrrrrrrr=r>x_stftspc mel_basisr.r/rrr logmelfilterbankBsF        rUc @seZdZdZdZeZedefddZ de de fddZ d e j de de j fd d Zed e j d e j dede j fddZede j defddZdS) LibrosaFbankzLibrosa fbank feature extractor Differs from Fbank extractor in that it uses librosa backend for stft and mel scale calculations. It can be easily configured to be compatible with existing speech-related projects that use librosa features. z librosa-fbankrcCs|jj|jjSr)configrrrrrr rHszLibrosaFbank.frame_shiftrcCs|jjSr)rWr)rrrrr feature_dimr"zLibrosaFbank.feature_dimsamplescCs&||jjksJt|fit|jSr)rWrrUr)rrYrrrr extractszLibrosaFbank.extract features_a features_benergy_scaling_factor_bc Cs&tttt||t|Sr)r6logrPrexp)r[r\r]rrr mixs zLibrosaFbank.mixfeaturescCsttt|Sr)floatr6sumr_)rarrr compute_energyszLibrosaFbank.compute_energyN)r%r&r'r(namer config_typepropertyr rHr)rXr6ndarrayrZr,rbr`rdrrrr rVs(  rV) dataclassesrrtypingrrnumpyr6lhotse.features.baserr lhotse.utilsrr r r r r rhr)rbr;r+rUrVrrrr sb    H