i!jddlmZddlmZmZddlZddlZddlmZddlZeeee fZ GddZ y))Path)TextUnionN)Tensorc eZdZdZededefdZededefdZ ede e e fde fdZdedefd Zdd ed effd Zded edefd Z ddedededefdZxZS)AudiouIAudio IO with on-the-fly resampling Parameters ---------- sample_rate: int Target sample rate. mono : int, optional Convert multi-channel to mono. Defaults to True. Usage ----- >>> audio = Audio(sample_rate=16000) >>> samples = audio("/path/to/audio.wav") # on-the-fly resampling >>> original_sample_rate = 44100 >>> two_seconds_stereo = torch.rand(2, 2 * original_sample_rate) >>> samples = audio({"samples": two_seconds_stereo, "sample_rate": original_sample_rate}) >>> assert samples.shape[1] == 2 * 16000 filereturnct|tr{d|vrg|d}t|jdk7s|jd|jdkDr t d|j dd}| t dyd |vryt d y) Nsamplesrz='samples' must be provided as a (channel, time) torch.Tensor. sample_ratez4'samples' must be provided with their 'sample_rate'.Taudioz1either 'audio' or 'samples' key must be provided.) isinstancedictlenshape ValueErrorget)r r rs s/Volumes/fast/ai/experiments/voice-extract-mac/.venv/lib/python3.12/site-packages/torch_audiomentations/utils/io.pyis_validzAudio.is_valid.s dD !D y/w}}%*gmmA.>qAQ.Q$W#hh}d; &$ND!!TUUr cp|jjddj}||dzz S)aPower-normalize samples Parameters ---------- samples : (..., time) Tensor Single (or multichannel) samples or batch of samples Returns ------- samples: (..., time) Tensor Power-normalized samples Tdimkeepdimg:0yE>)squaremeansqrt)r rmss r rms_normalizezAudio.rms_normalizeHs8nn##D#9>>@#*%%r file_pathctjt|}t|tur!|\}}|j }|j }||fS|j}|j}||fS)z"Return (num_samples, sample_rate).) torchaudioinfostrtypetuplelengthrate num_framesr)r$r'siei num_samplesrs rget_audio_metadatazAudio.get_audio_metadataYsms9~. : FB))K''KK''//K**KK''rc|j|t|tr4d|vr|djd}|d}n,|j |d\}}n|j |\}}t j ||jz|z S)zTNumber of samples (in target sample rate) :param file: audio file r rrr)rrrrr1mathceilr)selfr r0rs rget_num_sampleszAudio.get_num_sampleshs d dD !D "9o33A6 "=1 ,0+B+B4=+Q( [(,'>'>t'D $Kyyt'7'77+EFFrrmonoc>t|||_||_y)N)super__init__rr7)r5rr7 __class__s rr:zAudio.__init__s & rc|jr%|jddkDr|jdd}|j|k7r+tj j |||j}|S)a2Downmix and resample Parameters ---------- samples : (channel, time) Tensor Samples. sample_rate : int Original sample rate. Returns ------- samples : (channel, time) Tensor Remixed and resampled samples rrTr)r7rr rr& functionalresample)r5r rs rdownmix_and_resamplezAudio.downmix_and_resamplesg" 99q)A-llq$l7G   { * ++44d&6&6Gr sample_offsetr0c|j|d}t|tred|vr,|d}|d}|jd}|j dd}nVt |d}|j |\}}|j dd}n!t |}|j |\}}d}t||z|jz } ||| z } nt||z|jz } | | z|kDr t| tj| | \} } n|dd| | | zf} || |dz |ddf} |j| |} |o| jd |kDr | ddd|f} | S| jd |kr>|| jd z }tj j"j%| d |f} | S#t$r tdwxYw) a Parameters ---------- file : AudioFile Audio file. sample_offset : int, optional Start loading at this `sample_offset` sample. Defaults ot 0. num_samples : int, optional Load that many samples. Defaults to load up to the end of the file. Returns ------- samples : (time, channel) torch.Tensor Samples Nr rrchannelr) frame_offsetr-z~It looks like you are using an unsupported version of torchaudio. If you have 0.6 or older, please upgrade to a newer version.rr)rrrrrr(r1roundrrr&load TypeError Exceptionr?torchnnr=pad)r5r r@r0original_samplesoriginal_sample_rateoriginal_total_num_samplesrB audio_pathoriginal_sample_offsetoriginal_num_samples original_data_resultdiffs r__call__zAudio.__call__sK* d dD !D #' ? '+M':$-=-C-CA-F*((9d3!g/ ++J7.(((9d3TJ?C?V?V@ < &(<G!& 0 043C3C C"   #=@V#V #(22T5E5EE$  "$8 8;U U,   # #-??!73$ q-),BEY,YYYM  )'A+*?*BCM**=:NO  "||B+-rbs9  $d" #  hhr