mi%dZddlZddlmZddlZddlmZddlm Z ddl m Z Gdd ejjZ dd ejd ed ed ededeedefdZdZddZddZy)a  Differentiable, Pytorch based resampling. Implementation of Julius O. Smith algorithm for resampling. See https://ccrma.stanford.edu/~jos/resample/ for details. This implementation is specially optimized for when new_sr / old_sr is a fraction with a small numerator and denominator when removing the gcd (e.g. new_sr = 700, old_sr = 500). Very similar to [bmcfee/resampy](https://github.com/bmcfee/resampy) except this implementation is optimized for the case mentioned before, while resampy is slower but more general. N)Optional) functional)sinc simple_reprc reZdZdZd dedededeffd ZdZddejd e ed e fd Z d Z xZS) ResampleFracz? Resampling from the sample rate `old_sr` to `new_sr`. old_srnew_srzerosrolloffct|t|trt|ts t dt j ||}||z|_||z|_||_ ||_ |jy)a Args: old_sr (int): sample rate of the input signal x. new_sr (int): sample rate of the output. zeros (int): number of zero crossing to keep in the sinc filter. rolloff (float): use a lowpass filter that is `rolloff * new_sr / 2`, to ensure sufficient margin due to the imperfection of the FIR filter used. Lowering this value will reduce anti-aliasing, but will reduce some of the highest frequencies. Shape: - Input: `[*, T]` - Output: `[*, T']` with `T' = int(new_sr * T / old_sr) .. caution:: After dividing `old_sr` and `new_sr` by their GCD, both should be small for this implementation to be fast. >>> import torch >>> resample = ResampleFrac(4, 5) >>> x = torch.randn(1000) >>> print(len(resample(x))) 1250 z$old_sr and new_sr should be integersN) super__init__ isinstanceint ValueErrormathgcdr r r r _init_kernels)selfr r r rr __class__s I/var/www/stems/demucs_env/lib/python3.12/site-packages/julius/resample.pyrzResampleFrac.__init__so6 &#&j.ECD Dhhvv&m m    c|j|jk(ryg}t|j|j}||jz}t j |j |jz|z |_tj|j |j|jzj}t|jD]}| |jz ||jz z|z}|j|j |j }|tjz}tj||j z dz dz}t||z}|j!|j#|j%||j'dtj(|j+|jddy)Nkernelr)r r minrrceilr _widthtorcharangefloatrangeclamp_picosrdiv_sumappendregister_bufferstackview)rkernelssridxitwindowrs rrzResampleFrac._init_kernelsCsj ;;$++ %  dkk * dll&ii T[[ 82 => llDKK<t{{)BCIIKt{{# #ADKK#dkk/1R7A$**djj1A LAYYq|A~.1F!Wv%F KK % NN6 " # Xu{{7';'@'@aQS'TUrx output_lengthfullcR|j|jk(r|S|j}|jd}|jd|}t j |dddf|j |j |jzfd}t j||j|j}|jddjt|dddgz}tj|j|z|jz }tj|j} tj|j} ||r| n| } n:|dks|| kDrt!d | tj"|} |r t!d |d d| fS) a Resample x. Args: x (Tensor): signal to resample, time should be the last dimension output_length (None or int): This can be set to the desired output length (last dimension). Allowed values are between 0 and ceil(length * new_sr / old_sr). When None (default) is specified, the floored output length will be used. In order to select the largest possible size, use the `full` argument. full (bool): return the longest possible output from the input. This can be useful if you chain resampling operations, and want to give the `output_length` only for the last one, while passing `full=True` to all the other ones. rN replicate)mode)striderrrz$output_length must be between 0 and z0You cannot pass both full=True and output_length.)r r shapereshapeFpadr"conv1dr transposelistr# as_tensorr!longfloorrtensor) rr6r7r8r=lengthysyfloat_output_lengthmax_output_lengthdefault_output_lengthapplied_output_lengths rforwardzResampleFrac.forwardrsu ;;$++ %H IIb& ! EE!AtG*t{{DKK$++,EF[ Y XXaT[[ 9 LLA  & &tE#2J'72$'> ?#oodkkF.BT[[.PQ!JJ':;@@B % ,? @ E E G  9=$5CX ! Q -2C"CCDUCVWX X$)LL$? ! !STT,,,,--rct|S)Nr)rs r__repr__zResampleFrac.__repr__s 4  r)= ףp=?)NF)__name__ __module__ __qualname____doc__rr%rrr#TensorrboolrOrQ __classcell__)rs@rr r sZ$s$C$$5$L-V^#.#.hsm#.RV#.J!rr r6r r r rr7r8cLt||||j||||S)aR Functional version of `ResampleFrac`, refer to its documentation for more information. ..warning:: If you call repeatidly this functions with the same sample rates, then the resampling kernel will be recomputed everytime. For best performance, you should use and cache an instance of `ResampleFrac`. )r to)r6r r r rr7r8s r resample_fracr]s, ><w 7 : :1 =aPT UUrctjd|zdzd}|ddd}tj| dz|dz d|z}|tjz}t ||zj ddd}|S)NrF)periodicr?r)r# hann_windowlinspacerr(rr/)r winwinoddr4rs r_kernel_upsample2_downsample2rfs|   AIME :C AYF v|US[!e)sABw G L L Ze ZUY ZC QH"%A 166 5 " rc |jddzdk7rtj|d}|ddddf}|ddddf}|j^}}t|j |}|tj |j dd|||dddfj g||z}|j g|djd S) a Downsample x by a factor of two. The output length is half of the input, ceiled. Args: x (Tensor): signal to downsample, time should be the last dimension zeros (int): number of zero crossing to keep in the sinc filter. This function is kept only for reference, you should use the more generic `resample_frac` one. This function does not perform anti-aliasing filtering. rrr)rr.Nrrhra)r=r?r@rfr\rAr/mul)r6r xevenxoddrkrlrrms r _downsample2rss wwr{Q! EE!V  c3Q3hKE S!$Q$Yr{s   $!588??!F5:DI VU\\ V3 V V V,1 V!)# V=A V"")r