xi_@ dZddlZddlZddlZddlZddlZddlmZddlm Z m Z m Z m Z ddl Z ddlZddlZddlmZdde e j&ej(fdede e j&ej(ffd ZGd d Zdefd Zdefd ZdefdZ dde e j&ej(fde edefdZedZ dde e j&ej(fde eefdede ededef dZ y)z Audio saving and transcoding utility module Independent audio file operations outside of handler, supporting: - Save audio tensor/numpy to files (default FLAC format, fast) - Format conversion (FLAC/WAV/MP3) - Batch processing N)Path)UnionOptionalListTuple)logger audio_data target_dbreturnc\t|tjr|j}d}n|j }d}|r)tj tj |}n(tj tj |}|dkr|Sd|dz z}||z }||z}|S)a  Apply peak normalization to audio data. Args: audio_data: Audio data as torch.Tensor or numpy.ndarray target_db: Target peak level in dB (default: -1.0) Returns: Normalized audio data in the same format as input TFgư> g4@) isinstancetorchTensorclonecopymaxabsnp)r r audio is_tensorpeak target_ampgains 2/mnt/workspace/ACE-Step-1.5/acestep/audio_utils.pynormalize_audiors*ell+  " ! yy5)*vvbffUm$ d{ D()J  D DLE Lc8eZdZdZddefdZ ddeeje jfdeee fde de ed ed ef d Z dd eee fdeee fd eded ef dZ ddeeejejfdeee fdede de ed ed eefdZy) AudioSaverz*Audio saving and transcoding utility classdefault_formatc|j|_|jdvr!tjd|dd|_yy)z Initialize audio saver Args: default_format: Default save format ('flac', 'wav', 'mp3', 'wav32', 'opus', 'aac') flacwavmp3wav32opusaacUnsupported format z, using 'flac'r#N)lowerr rwarning)selfr s r__init__zAudioSaver.__init__EsI-224   &T T NN00@O P"(D  UrNr output_path sample_rateformatchannels_firstr c |xs |jj}|dvr1tjd|d|j|j}t |}|dk(rdnd|}|j jdvr|j |}nW|dk(r/|j jdk(r|j d}n#|d k(r|j jd k(r t|tjr|r$tj|j}ntj|j}|jd k(r|jd |jd kDrk|j}n^|j!j}|s>|jd k(r+|jd |jd kDr |j}|j#} |dvr$t%j&t)|||ddn|dvr|dk(rk d dl}|j-d d j/} |j1t)|| |ddtj2d|d|dt)|St%j&t)|||ddn"t%j&t)|||dtj2d|d|d |dt)|S#t4$r%} tj6d| dd}Yd} ~ d} ~ wwxYw#t4$r} d dl}|j-d d j/} |dk(rd} d} n|j9} d} |j1t)|| || | !tj2d"|d|d |dt)|cYd} ~ S#t4$r!} tj6d#| d$| d} ~ wwxYwd} ~ wwxYw)%a  Save audio data to file Args: audio_data: Audio data, torch.Tensor [channels, samples] or numpy.ndarray output_path: Output file path (extension can be omitted) sample_rate: Sample rate format: Audio format ('flac', 'wav', 'mp3', 'wav32', 'opus', 'aac'), defaults to default_format channels_first: If True, tensor format is [channels, samples], else [samples, channels] Returns: Actual saved file path r"r)z, using r&.wav.)z.flacr3z.mp3z.opusz.aac.m4az.wav32r(r5r)r%r'r(Tffmpeg)r1backend)r#r$r&NFLOATWAV)subtyper0z[AudioSaver] Saved audio to z (wav32, zHz)zFailed to save wav32: z, falling back to standard wavr$ soundfile)r1z (z, )r0r<z/[AudioSaver] Fallback soundfile Saved audio to z#[AudioSaver] Failed to save audio: z -> Fallback failed: )r r*rr+rsuffix with_suffixrrndarrayr from_numpyfloatdimshapeTcpu contiguous torchaudiosavestrr= transposenumpywritedebug Exceptionerrorupper)r,r r.r/r0r1ext audio_tensorsfaudio_npe sf_formatr<inner_es r save_audiozAudioSaver.save_audioQs*/D//668 G G NN0ATAT@UV W((F;' ')f6(|    # # %-_ _%11#6K w ;#5#5#;#;#=#I&226:[ u_!3!3!9!9!;v!E  j"** -$// ;AAC  %// ;AAC ##%*|/A/A!/D|GYGYZ[G\/\$0NN\&>>+113L!l&6&6&8A&=%%a(<+=+=a+@@#/>>L$..0 B // $ #'$ 33W$ '.$0#9#9!Q#?#E#E#G[!18[RYbgh 'CK=PYZeYffi%jk";//  $ #''  $ #'  LL7 }BvhbQ\P]]`a b{# #-%' '=aS@^%_`!&'0  &'11!Q7==?W$ %I%G & I"G[)8[\cd N{m[]^d]eeghsgttwxy;''  B1#EZ[bZcde ! sc1L>3A)L A/L> L;L61L>6L;;L>> PBOP PO;;PPP input_path output_format remove_inputc0t|}t|}|jstd|tjt |\}}|j ||||d}|r(|jtjd||S)aa Convert audio format Args: input_path: Input audio file path output_path: Output audio file path output_format: Target format ('flac', 'wav', 'mp3', 'wav32', 'opus', 'aac') remove_input: Whether to delete input file Returns: Output file path zInput file not found: Tr/r0r1z![AudioSaver] Removed input file: ) rexistsFileNotFoundErrorrHloadrJrYunlinkrrN)r,rZr.r[r\rSr/s r convert_audiozAudioSaver.convert_audios&*% ;'   "#&L<#8#8~~HI|4499;< $$\2HHV$   77>>) $   i ( WW]]9 %  Y677>>%(%%e, E %&! !w$! . " ' !  )=@$-8   4"4'&  s* F-(F!> F-!F* &F-- F98F9c|y t|trtjj |rGt |d5}t j|jjcdddSt j|jdjSt|drEt jt|jjdjSt jt|jdjS#1swYxYw#t$r>t jt|jdjcYSwxYw)z Get hash identifier for an audio file. Args: audio_file: Path to audio file (str) or file-like object Returns: Hash string or empty string Nrrutf-8name)rrJrrr_rrrrrencodehasattrrrO) audio_filers rget_audio_file_hashros K j# &ww~~j)*d+q">>!&&(3==?,+>>*"3"3G"<=GGI I Z (>>#joo"6"="=g"FGQQS S~~c*o44W=>HHJJ ,+ K~~c*o44W=>HHJJKs=;E1D91 E;1E-AE>:E9E>EAF  F c tj|dd}tj|j d}|j }|ddd|ddd|dd d|d d d|d d  }|S) z Generate deterministic UUID from generation parameters. Same parameters will always generate the same UUID. Args: params_dict: Dictionary of parameters Returns: UUID string TF) sort_keys ensure_asciirr-  )jsondumpsrrrr) params_dict params_jsonrhash_hexuuid_strs rgenerate_uuid_from_paramsrs**[DuMK~~k009:H!!#H1Q-(1R.!18Br?2C1XbQS_DUUVW_`bceWfVghH OrseedcRt|tjr|jj }n|}t j |jj}|8|d|}t j |jjS|S)z Generate UUID from audio data (for caching/deduplication) Args: audio_data: Audio data seed: Optional seed value Returns: UUID string rk) rrrrFrLrrtobytesrr)r rrU data_hashcombineds rgenerate_uuid_from_audio_datars*ell+>>#))+x//12<<>I [$(~~hoo/0::<< rr#)r r.r/r0r1c4tj|||||S)a3 Convenience function: save audio (using default configuration) Args: audio_data: Audio data output_path: Output path sample_rate: Sample rate format: Format (default flac) channels_first: Tensor format flag Returns: Saved file path )_default_saverrY)r r.r/r0r1s rrYrYs"(  $ $Kfn r)g)Nrw)!r|iorr subprocessrpathlibrtypingrrrrrrLrrHlogururrr@rBrrrJrrrr}rrr~rYrrrrsW // &ellBJJ&> ?&E&]bchcocoqsq{q{c{]|&Tnnb95#95xKsK4c*ellBJJ./ 3- >62   ellBJJ./sDy! SM     r