o ϯiV/@sddlZddlmZddlZddlmmZddlmZddlm Z ddlmZGdddej Z Gdd d ej Z Gd d d ej Z Gd d d ej ZGdddej ZdS)N)List) AudioSignal STFTParams)nncsBeZdZdZd dedeffdd Zded effd d ZZ S) L1LossaL1 Loss between AudioSignals. Defaults to comparing ``audio_data``, but any attribute of an AudioSignal can be used. Parameters ---------- attribute : str, optional Attribute of signal to compare, defaults to ``audio_data``. weight : float, optional Weight of this loss, defaults to 1.0. Implementation copied from: https://github.com/descriptinc/lyrebird-audiotools/blob/961786aa1a9d628cca0c0486e5885a457fe70c1a/audiotools/metrics/distance.py audio_data? attributeweightc s"||_||_tjdi|dS)N)r r super__init__)selfr r kwargs __class__r B/home/ubuntu/.local/lib/python3.10/site-packages/dacvae/nn/loss.pyrszL1Loss.__init__xycs0t|trt||j}t||j}t||S)a  Parameters ---------- x : AudioSignal Estimate AudioSignal y : AudioSignal Reference AudioSignal Returns ------- torch.Tensor L1 loss between AudioSignal attributes. ) isinstancergetattrr r forward)rrrrr rr!s   zL1Loss.forward)rr ) __name__ __module__ __qualname____doc__strfloatrrr __classcell__r r rrr src sTeZdZdZ     ddededed ed ef fd d Zd edefddZ Z S) SISDRLossa Computes the Scale-Invariant Source-to-Distortion Ratio between a batch of estimated and reference audio signals or aligned features. Parameters ---------- scaling : int, optional Whether to use scale-invariant (True) or signal-to-noise ratio (False), by default True reduction : str, optional How to reduce across the batch (either 'mean', 'sum', or none).], by default ' mean' zero_mean : int, optional Zero mean the references and estimates before computing the loss, by default True clip_min : int, optional The minimum possible loss value. Helps network to not focus on making already good examples better, by default None weight : float, optional Weight of this loss, defaults to 1.0. Implementation copied from: https://github.com/descriptinc/lyrebird-audiotools/blob/961786aa1a9d628cca0c0486e5885a457fe70c1a/audiotools/metrics/distance.py TmeanNr scaling reduction zero_meanclip_minr cs,||_||_||_||_||_tdSN)r"r#r$r%r r r)rr"r#r$r%r rr rrNs zSISDRLoss.__init__rrcCsfd}t|tr|j}|j}n|}|}|jd}||ddddd}||ddddd}|jrA|jddd}|jddd}nd}d}||} ||} | djdd |} | | jdd |} |j rk| |  dnd} | | }| |}|djdd }|djdd }d t |||}|j durt j||j d }|jd kr|}|S|jd kr|}|S)Ng:0yE>rT)dimkeepdim)r*i)minr!sum)rrrshapereshapepermuter$r!r.r" unsqueezetorchlog10r%clampr#)rrreps references estimatesnbmean_reference mean_estimate _references _estimatesreferences_projectionreferences_on_estimatesscalee_truee_ressignalnoisesdrr r rr]sF     zSISDRLoss.forward)Tr!TNr ) rrrrintrrrrrrr r rrr 5s&r cszeZdZdZddgedddddddf d eed ej d e d e d e de de de de ffdd Z dedefddZZS)MultiScaleSTFTLossaComputes the multi-scale STFT loss from [1]. Parameters ---------- window_lengths : List[int], optional Length of each window of each STFT, by default [2048, 512] loss_fn : typing.Callable, optional How to compare each loss, by default nn.L1Loss() clamp_eps : float, optional Clamp on the log magnitude, below, by default 1e-5 mag_weight : float, optional Weight of raw magnitude portion of loss, by default 1.0 log_weight : float, optional Weight of log magnitude portion of loss, by default 1.0 pow : float, optional Power to raise magnitude to before taking log, by default 2.0 weight : float, optional Weight of this loss, by default 1.0 match_stride : bool, optional Whether to match the stride of convolutional layers, by default False References ---------- 1. Engel, Jesse, Chenjie Gu, and Adam Roberts. "DDSP: Differentiable Digital Signal Processing." International Conference on Learning Representations. 2019. Implementation copied from: https://github.com/descriptinc/lyrebird-audiotools/blob/961786aa1a9d628cca0c0486e5885a457fe70c1a/audiotools/metrics/spectral.py h㈵>r @FNwindow_lengthsloss_fn clamp_eps mag_weight log_weightpowr match_stride window_typec sHtfdd|D|_||_||_||_||_||_||_dS)Nc g|] }t||ddqS) window_length hop_lengthrRrSr.0wrRrSr r z/MultiScaleSTFTLoss.__init__..) r r stft_paramsrMrPrOrNr rQ) rrLrMrNrOrPrQr rRrSrr\rrs   zMultiScaleSTFTLoss.__init__rrc Csd}|jD]D}||j|j|j||j|j|j||j||j|j  |j  |j|j  |j  7}||j ||j|j7}q|S)a?Computes multi-scale STFT between an estimate and a reference signal. Parameters ---------- x : AudioSignal Estimate signal y : AudioSignal Reference signal Returns ------- torch.Tensor Multi-scale STFT loss. ) r_stftrWrXrSrPrM magnituder5rNrQr4rO)rrrlosssr r rrs  zMultiScaleSTFTLoss.forwardrrrrrrrrFtypingCallablerboolrrrrrr r rrrGs>!  rGcseZdZdZddgddgedddddd d d gd d gd f d eed eedej de de de de de de dee dee de ffdd Z dedefddZZS)MelSpectrogramLossafCompute distance between mel spectrograms. Can be used in a multi-scale way. Parameters ---------- n_mels : List[int] Number of mels per STFT, by default [150, 80], window_lengths : List[int], optional Length of each window of each STFT, by default [2048, 512] loss_fn : typing.Callable, optional How to compare each loss, by default nn.L1Loss() clamp_eps : float, optional Clamp on the log magnitude, below, by default 1e-5 mag_weight : float, optional Weight of raw magnitude portion of loss, by default 1.0 log_weight : float, optional Weight of log magnitude portion of loss, by default 1.0 pow : float, optional Power to raise magnitude to before taking log, by default 2.0 weight : float, optional Weight of this loss, by default 1.0 match_stride : bool, optional Whether to match the stride of convolutional layers, by default False Implementation copied from: https://github.com/descriptinc/lyrebird-audiotools/blob/961786aa1a9d628cca0c0486e5885a457fe70c1a/audiotools/metrics/spectral.py PrHrIrJr rKFr`Nn_melsrLrMrNrOrPrQr rRmel_fminmel_fmaxrSc sZtfdd|D|_||_||_||_||_||_||_| |_ | |_ ||_ dS)NcrTrUrrYr\r rr]r^z/MelSpectrogramLoss.__init__..) r rr_rlrMrNrPrOr rmrnrQ) rrlrLrMrNrOrPrQr rRrmrnrSrr\rrs    zMelSpectrogramLoss.__init__rrc Csd}t|j|j|j|jD]Q\}}}}|j|j|jd}|j|f||d|} |j|f||d|} ||j | | |j  |j | |j  |j 7}||j| | | 7}q |S)a*Computes mel loss between an estimate and a reference signal. Parameters ---------- x : AudioSignal Estimate signal y : AudioSignal Reference signal Returns ------- torch.Tensor Mel loss. r`)rWrXrS)rmrn)ziprlrmrnr_rWrXrSmel_spectrogramrPrMr5rNrQr4rO) rrrrcrlfminfmaxrdrx_melsy_melsr r rr(s  zMelSpectrogramLoss.forwardrer r rrrisP     #rics8eZdZdZfddZddZddZdd ZZS) GANLossz Computes a discriminator loss, given a discriminator on generated waveforms/spectrograms compared to ground truth waveforms/spectrograms. Computes the loss for both the discriminator and the generator in separate functions. cst||_dSr&)r r discriminator)rrvrr rrTs  zGANLoss.__init__cCs ||j}||j}||fSr&)rvr)rfakereald_faked_realr r rrXs  zGANLoss.forwardcCsd|||\}}d}t||D]\}}|t|dd7}|td|dd7}q|S)Nrr(r)r')rclonedetachror3r!)rrwrxryrzloss_dx_fakex_realr r rdiscriminator_loss]s zGANLoss.discriminator_lossc Cs|||\}}d}|D]}|td|dd7}q d}tt|D]!}tt||dD]} |t||| ||| 7}q0q$||fS)Nrr'r(r))rr3r!rangelenFl1_lossr|) rrwrxryrzloss_gr~ loss_featureijr r rgenerator_lossfs&zGANLoss.generator_loss) rrrrrrrrrr r rrruLs   ru)rfrr3torch.nn.functionalr functionalr audiotoolsrrrModuler rGrirur r r rs    ([Yc