o rriH*@sdZddlmZddlZddlZddlZddlmZddl m Z ddl Z ddl mZddlmZdd lmZdd lmZmZejeeefZGd d d ejZGd ddZddZ     d!dejeefdejejefdededededededej edejfdd Z!dS)"z Code to apply a model to a mix. It will handle chunking with overlaps and inteprolation between chunks, as well as the "shift trick". )ThreadPoolExecutorN)nn) functional)Demucs)HDemucs)HTDemucs) center_trimDummyPoolExecutorcsheZdZ  d dejedejejejedejeffdd Ze defdd Z d d Z Z S) BagOfModelsNmodelsweightssegmentcstt|dks J|d|D]!}|jjksJ|jjks%J|jjks-J|dur4||_qj|_j|_j|_t||_ |durUfdd|D}nt|t|ks_J|D] }t|tjksnJqa||_ dS)a Represents a bag of models with specific weights. You should call `apply_model` rather than calling directly the forward here for optimal performance. Args: models (list[nn.Module]): list of Demucs/HDemucs models. weights (list[list[float]]): list of weights. If None, assumed to be all ones, otherwise it should be a list of N list (N number of models), each containing S floats (S number of sources). segment (None or float): overrides the `segment` attribute of each model (this is performed inplace, be careful is you reuse the models passed). rNcsg|] }ddjDqS)cSsg|]}dqS)?.0_rr@/home/ubuntu/.local/lib/python3.10/site-packages/demucs/apply.py <sz3BagOfModels.__init__...)sourcesrfirstrrr<sz(BagOfModels.__init__..) super__init__lenr samplerateaudio_channelsrr ModuleListr r )selfr r rotherweight __class__rrrs(   zBagOfModels.__init__returncCs2td}|jD]}t|trt|t|j}q|S)Ninf)floatr isinstancerminr)rmax_allowed_segmentmodelrrrr)Cs   zBagOfModels.max_allowed_segmentcCstd)NzCall `apply_model` on this.)NotImplementedError)rxrrrforwardKszBagOfModels.forward)NN) __name__ __module__ __qualname__tpListModelOptionalr&rpropertyr)r- __classcell__rrr"rr s'r c@s*eZdZd ddZeddZddZdS) TensorChunkrNcCs|jd}|dks J||ksJ|dur||}nt|||}t|tr1|j|_||j|_n||_||_||_|j|_dS)Nr)shaper(r'r7tensoroffsetlengthdevice)rr:r;r< total_lengthrrrrPs      zTensorChunk.__init__cCst|jj}|j|d<|S)Nr8)listr:r9r<)rr9rrrr9cs  zTensorChunk.shapec Cs||j}|jjd}|dksJ|j|d}||}td|}t||}||}||} t|jd||f|| f} | jd|ksFJ| S)Nr8r.)r<r:r9r;maxr(Fpad) r target_lengthdeltar>startend correct_start correct_endpad_left pad_rightoutrrrpaddedis     zTensorChunk.padded)rN)r.r/r0rr5r9rMrrrrr7Os    r7cCs&t|tr|St|tjsJt|S)N)r'r7thTensor)tensor_or_chunkrrr tensor_chunk|s rQT?rFr*mixshiftssplitoverlaptransition_powerprogress num_workersrr$c ) Cs|dur|j}nt|}| dur"|dkr|jdkrt|} nt} ||||||| | d} t|trd} dgt|j} t |j |j D]H\}}t t |j}||t||fi| }||t|D]\}}|dd|ddddf|9<| ||7<qg| |7} ~qCt| tjsJt| jdD]}| dd|ddddf| |<q| S||||dksJd|j\}}}|r*d| d<td |j}t|}t|tsJ||d |}d}t|D]'}td|}t|||||}t||fi| }||d ||df7}q||}t|tjs(J|S|r,d | d <tj|t|j|||jd}tj||jd}| durO|j} | durY| dks[Jt|j| }td||}td||} tt ||jd}!t!tj"d|d d|dtj"||d dd|dg}"t|"|ksJ|"|"#|}"g}#| D]}t|||}$| j$t||$fi| }%|#%|%|f||7}q|rt&j&|#|!ddd}#|#D]:\}%}|%'}&|&jd}'|d |||f|"d|'|&|j7<|||||"d|'|j7<q|(dksJ||}t|tjs*J|St|t)r?| dur?t| |j}(nt*|drK|+|}(n|}(t|}t|tsYJ||(|}t, ||}Wdn 1suwYt|tjsJt-||S)a Apply model to a given mixture. Args: shifts (int): if > 0, will shift in time `mix` by a random amount between 0 and 0.5 sec and apply the oppositve shift to the output. This is repeated `shifts` time and all predictions are averaged. This effectively makes the model time equivariant and improves SDR by up to 0.2 points. split (bool): if True, the input will be broken down in 8 seconds extracts and predictions will be performed individually on each and concatenated. Useful for model with large memory footprint like Tasnet. progress (bool): if True, show a progress bar (requires split=True) device (torch.device, str, or None): if provided, device on which to execute the computation, otherwise `mix.device` is assumed. When `device` is different from `mix.device`, only local computations will be on `device`, while the entire tracks will be stored on `mix.device`. num_workers (int): if non zero, device is 'cpu', how many threads to use in parallel. segment (float or None): override the model segment parameter. Nrcpu)rTrUrVrWrXr=poolrgrz-transition_power < 1 leads to weird behavior.rTg?r@.FrU)r=z.2fr8xseconds) unit_scalencolsunit valid_length).r=rNtyperr r'r rrzipr r nextiter parametersto apply_model enumeraterOranger9evalintrrQr7rMrandomrandintzerosrr&formatcatarangerAsubmitappendtqdmresultr(rhasattrrano_gradr ))r*rSrTrUrVrWrXr=rYrr[kwargs estimatestotals sub_model model_weightsoriginal_model_devicerLk inst_weightbatchchannelsr< max_shift padded_mixrr;shifted shifted_out sum_weightsegment_lengthstrideoffsetsscaler!futureschunkfuture chunk_out chunk_lengthrarrrrhs     $*          ,     rh) rTrRrFNrNN)"__doc__concurrent.futuresrrmtypingr1torchrNrtorch.nnrrBrudemucsrhdemucsrhtdemucsrutilsr r Unionr3Moduler r7rQrOrlboolr&r4rhrrrrsP      4-