:jQ ddlZddlmZddlmZmZddlmZmZddl Z ddl Z ddl m cmZddlmZdZdZd Zd ZeezZeeeZed zZeeeZeeeZefd ed efdZefdddedefdZeddede j>fdZ ddeee jBe j>fdededeeee jDffdZ#y)N) lru_cache)CalledProcessErrorrun)OptionalUnion) exact_divi>ifilesrc|ddddd|dddd d d d t|d g} t|ddj}tj|tjjjtjdz S#t$r,}t d|j j |d}~wwxYw)a? Open an audio file and read as mono waveform, resampling as necessary Parameters ---------- file: str The audio file to open sr: int The sample rate to resample the audio if necessary Returns ------- A NumPy array containing the audio waveform, in float32 dtype. ffmpegz-nostdinz-threads0z-iz-fs16lez-ac1z-acodec pcm_s16lez-ar-T)capture_outputcheckzFailed to load audio: Ng@) strrstdoutr RuntimeErrorstderrdecodenp frombufferint16flattenastypefloat32)r rcmdoutes b/home/mcse/projects/srt_converter/srt-converter-venv/lib/python3.12/site-packages/whisper/audio.py load_audior's* C d g s; s2w  CP#d$7>> ==bhh ' / / 1 8 8 Dw NN P3AHHOO4E3FGHaOPsB B;'B66B;)axislengthr)c tj|r|j||kDr2|j|tj||j }|j||krZdg|j z}d||j|z f||<tj||dddDcgc] }|D]}| c}}}|S|j||kDr|jt||}|j||kr=dg|j z}d||j|z f||<tj||}|Scc}}w)zO Pad or trim the audio array to N_SAMPLES, as expected by the encoder. )device)dimindex)rrrNr()indicesr)) torch is_tensorshape index_selectaranger,ndimFpadtakeranger)arrayr*r) pad_widthssizesr7s r& pad_or_trimr=AsC u ;;t v %&& VELL I'E ;;t v % EJJ.J !6EKK,=#=>Jt EE%:dd3C!U%u!U#!U#!UVE L ;;t v %JJuV}4J@E ;;t v % EJJ.J !6EKK,=#=>Jt FF5*-E L"Vs.E )maxsizen_melsreturncP|dvs Jd|tjjtjjtdd}t j |d5}tj|d|j|cdddS#1swYyxYw) ad load the mel filterbank matrix for projecting STFT into a Mel spectrogram. Allows decoupling librosa dependency; saved using: np.savez_compressed( "mel_filters.npz", mel_80=librosa.filters.mel(sr=16000, n_fft=400, n_mels=80), mel_128=librosa.filters.mel(sr=16000, n_fft=400, n_mels=128), ) >PzUnsupported n_mels: assetszmel_filters.npzF) allow_picklemel_N) ospathjoindirname__file__rloadr0 from_numpyto)r,r? filters_pathfs r& mel_filtersrQ[s Y ?"6vh ?? 77<< 98EVWL E 2?aD/ 2366v>???s (*BB%audiopaddingr,ctj|s0t|tr t |}tj |}||j |}|dkDrtj|d|f}tjtj |j}tj|tt|d}|dddfjdz}t|j|}||z}tj |d j#} tj$| | j'd z } | d zd z } | S) a} Compute the log-Mel spectrogram of Parameters ---------- audio: Union[str, np.ndarray, torch.Tensor], shape = (*) The path to audio or either a NumPy array or Tensor containing the audio waveform in 16 kHz n_mels: int The number of Mel-frequency filters, only 80 and 128 are supported padding: int Number of zero samples to pad to the right device: Optional[Union[str, torch.device]] If given, the audio tensor is moved to this device before STFT Returns ------- torch.Tensor, shape = (n_mels, n_frames) A Tensor that contains the Mel spectrogram NrT)windowreturn_complex.r(r g|=)ming @g@)r0r1 isinstancerr'rMrNr6r7 hann_windowN_FFTr,stft HOP_LENGTHabsrQclamplog10maximummax) rRr?rSr,rUr[ magnitudesfiltersmel_speclog_specs r&log_mel_spectrogramrfns8 ??5 ! eS !u%E  '  {ea\*   u % ( ( 6F ::eUJvd SDc3B3h##%*J%,,/G#H{{8/557H}}Xx||~';r{s2 ."      ; & Y + 1nk:6k+>?%0%OS%Oc%OP&/rs4 4?? ??(15 / bjj%,,. // // U3 ,- . /rg