SܶiQ ddlZddlmZddlmZmZddlmZmZddl Z ddl Z ddl m cmZddlmZdZdZd Zd ZeezZeeeZed zZeeeZeeeZefd ed efdZefdddedefdZeddede jfdZ ddeee j!e jfdededeeee j"ffdZ#dS)N) lru_cache)CalledProcessErrorrun)OptionalUnion) exact_divi>ifilesrcddddd|dddd d d d t|d g} t|ddj}n<#t$r/}t d|j|d}~wwxYwtj|tj  tj dz S)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 P/root/projects/openclaw-proxy/venv/lib/python3.11/site-packages/whisper/audio.py load_audior's* C d g s; s2ww  CP#d$777> PPPGAHOO4E4EGGHHaOP =bh ' ' / / 1 1 8 8 D Dw NNs8 A1*A,,A1)axislengthr)cZtj|r|j||kr0||tj||j}|j||krHdg|jz}d||j|z f||<tj|d|dddD}n{|j||kr$| t||}|j||kr5dg|jz}d||j|z f||<tj||}|S) zO Pad or trim the audio array to N_SAMPLES, as expected by the encoder. )device)dimindex)rrrcg|] }|D]}| Sr0).0sizespads r& zpad_or_trim..Ns%!U!U!U%u!U!U#!U!U!U!UNr()indicesr)) torch is_tensorshape index_selectaranger,ndimFr3takeranger)arrayr*r) pad_widthss r& pad_or_trimrBAs? u. ;t v % %&& VEL I I I'E ;t v % % EJ.J !6EK,=#=>Jt E%!U!U:ddd3C!U!U!UVVE ;t v % %JJuV}}4J@@E ;t v % % EJ.J !6EK,=#=>Jt F5*--E Lr5)maxsizen_melsreturncj|dvs Jd|tjtjtdd}t j|d5}tj|d| |cdddS#1swxYwYdS) 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__rloadr7 from_numpyto)r,rD filters_pathfs r& mel_filtersrV[s Y    ?v ? ?   7<< 9 98EVWWL E 2 2 2?a/// 23366v>>??????????????????s+0B((B,/B,rGaudiopaddingr,ctj|s8t|trt |}tj|}|||}|dkrtj|d|f}tj t|j }tj |tt|d}|dddfdz}t|j |}||z}tj|d } tj| | 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@)r7r8 isinstancerr'rRrSr=r3 hann_windowN_FFTr,stft HOP_LENGTHabsrVclamplog10maximummax) rWrDrXr,rZr` magnitudesfiltersmel_speclog_specs r&log_mel_spectrogramrknsB8 ?5 ! !( eS ! ! &u%%E ''   {{ea\**  u % % ( ( 6 6F :eUJvd S S SDc3B3h##%%*J%,//G#H{8///5577H}Xx||~~';<r~s ........""""""""      ; & 9Y + + 1nIk:66Ik+>??%0%O%OS%Oc%O%O%O%OP&/rs4 4?? ????(15 // bj%,. // // U3 ,- . //////r5