ihddlmZddlmZddlmZddlmZmZm Z m Z ddl m Z ddl mZmZmZmZddlZddlZddlmZddlmZdd lmZdd lmZdd Zd fd edeedeefdZde defdZ!dede fdZ" ddejFdee efd edeedeedeedeejFdejFfdZ$y))CounterFraction)reduce)chaincountislicerepeat)log2)CallableListOptionalUnionN)version)primespadc #(K|r"tt|j\}d}t|}t t |t ||}t||kry tfd|Dttt||tt|t|D]\}}|||k7sny||dz}tt|||t |||dD] \}}|||< w)NcJtjtt||SN)r from_iterablemapr )ics k/Volumes/fast/ai/experiments/voice-extract-mac/.venv/lib/python3.12/site-packages/torch_pitch_shift/main.pyz2_combinations_without_repetition..sU((VQ):;c3(K|] }| ywr).0rvaluess r z3_combinations_without_repetition..s/w!F1Iws) zipritemslenlistr rtuplereversedrange) riterabler!countsfnindicesrjs ` r _combinations_without_repetitionr2sgh/5578;A F A6!EGV,a01G 7|a /w///q*AhuQx.@(6BR,STDAqqzQU  AJNa QuQx%<=DAqGAJ> s C DADc(|dk\xr |dkxr|dk7S)Ng?r#r)xs rrr(sa3h.L16.La1f.Lr sample_rate conditionreturnc pt}tj|}g}tdt |dzD]7}|j t ||Dcgc]}td|c}9|D]/}|D](}t||}||s|j|*1t|Scc}w)a Search for pitch-shift targets that can be computed quickly for a given sample rate. Parameters ---------- sample_rate: int The sample rate of an audio clip. condition: Callable [optional] A function to validate fast shift ratios. Default is `lambda x: x >= 0.5 and x <= 2 and x != 1` (between -1 and +1 octaves). Returns ------- output: List[Fraction] A list of fast pitch-shift target ratios r#)r,c ||zSrr)r5ys rrz!get_fast_shifts..@sAEr) setrfactorsr*r&extendr2rraddr') r6r7 fast_shiftsr=productsrr5r1r.s rget_fast_shiftsrB&s(%Knn[)GH 1c'lQ& ':!gN NA)1-N  (AAA|"   sB3 semitonesc$td|dz zS)z Convert semitonal shifts into ratios. Parameters ---------- semitones: float The number of semitones for a desired shift. Returns ------- output: Fraction A Fraction indicating a pitch shift ratio @(@r)rCs rsemitones_to_ratiorGLs CI,- ..rratioc0tdt|zS)z Convert rational shifts to semitones. Parameters ---------- ratio: Fraction The ratio for a desired shift. Returns ------- output: float The magnitude of a pitch shift in semitones rF)floatr )rHs rratio_to_semitonesrK]s U # $$rinputshiftbins_per_octaven_fft hop_lengthwindowc|s|dz}|s|dz}|tj|}|j|j}|j\}}} t |t sdt||z z}tj|t||z j|j} |} | j||z| } tjtjtjdk\} tj | ||| |d} tj"td|z | jd | j|j} | | } tj$| d ||| } | | } ~ ~ | jd|jd k\r| ddd|jd f} n.t'| d |jd | jdz d d f } | j||| } | S)a Shift the pitch of a batch of waveforms by a given amount. Parameters ---------- input: torch.Tensor [shape=(batch_size, channels, samples)] Input audio clips of shape (batch_size, channels, samples) shift: float OR Fraction `float`: Amount to pitch-shift in # of bins. (1 bin == 1 semitone if `bins_per_octave` == 12) `Fraction`: A `fractions.Fraction` object indicating the shift ratio. Usually an element in `get_fast_shifts()`. sample_rate: int The sample rate of the input audio clips. bins_per_octave: int [optional] Number of bins per octave. Default is 12. n_fft: int [optional] Size of FFT. Default is `sample_rate // 64`. hop_length: int [optional] Size of hop length. Default is `n_fft // 32`. window: torch.Tensor [optional] A window tensor for the STFT. Default is a tensor of ones. Returns ------- output: torch.Tensor [shape=(batch_size, channels, samples)] The pitch-shifted batch of audio clips @ NrEz0.11.0)return_complexrQ)N.r#r4) fixed_raten_freqrPr)rQr)torchonestodeviceshape isinstancerrJTResampleintreshaperparse torchaudio __version__stft TimeStretchistftr)rLrMr6rNrOrPrQ batch_sizechannelssamples resampleroutputv011 stretchers r pitch_shiftronsH r! b[  ~E" YYu|| $F$)KK!J' eX &e 67 ;K%,?(@ADDU\\RI F ^^J17 ;F ==// 0GMM(4K KD ZZz$v VF U#FLLO bv F [[E:f EF v F9 ||A%++a.(-ekk!n--.V!U[[^fll1o%Eq!!LM ^^J' :F Mr)NNN) rrN)% collectionsr fractionsr functoolsr itertoolsrrr r mathr typingr r rrrXrctorchaudio.transforms transformsr^ packagingrprimePyrtorch.nn.functionalrr2r`rBrJrGrKTensorrorrrr}s2222 !#.%M##!# (^#L/%/H/"%h%5%*&( !%)B <<B  !BBc] B C= B  B U\\ "B \\Br