.`ib ddlmZddlmZddlmZddlZddlmZ ddl Z ddl m Z  ddl Z n#e$re dZ YnwxYw ddlmZn)#e$r!e ddZYnwxYwGd d eeZeGd d Zed ejZedZde jeje jzdede jeje jzfdZde jejdedede jejfdZde jejdedefdZGddZ dS)) dataclass)Enum)LiteralN)PlaceholderModulelibrosascipysignalc"eZdZdZdZdZdZdZdS)ChannelReductionz8Method to reduce multi-channel audio to target channels.meanfirstmaxsumN)__name__ __module__ __qualname____doc__MEANFIRSTMAXSUMi/home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/vllm/multimodal/audio.pyr r s(BB D E C CCCrr cheZdZUdZdZedzed<ejZ eed<e de fdZ de fdZdS) AudioSpecaSpecification for target audio format. This dataclass defines the expected audio format for a model's feature extractor. It is used to normalize audio data before processing. Attributes: target_channels: Number of output channels. None means passthrough (no normalization). 1 = mono, 2 = stereo, etc. channel_reduction: Method to reduce channels when input has more channels than target. Only used when reducing channels. Ntarget_channelschannel_reductionreturnc|jduS)z&Whether audio normalization is needed.Nrselfs rneeds_normalizationzAudioSpec.needs_normalization5s#4//rcD|jdSd|jd|jjdS)NzAudioSpec(passthrough)zAudioSpec(channels=z , reduction=))rrvaluer#s r__repr__zAudioSpec.__repr__:sB   '++ 9$"6 9 9/5 9 9 9 r)rrrrrint__annotations__r rrpropertyboolr%strr)rrrrr$s  #$OS4Z###*:*?'??? 0T000X0 #      rrr)rrr"audiospecr c:|js|S|jdkr%|jdkr|Std|jd|jdkrtd|jd|jd|jdkr(t |t jr|jn|j}|jd}||jkr|S||jkrtd|d |jt |t j}|jdkr|j tj kr/|rt j |d n| d }n|j tj kr |d}n|j tjkr4|rt j|d n|d j}n[|j tjkr/|rt j|d n|d }ntd |j |S|d |jS)aKNormalize audio to the specified format. This function handles channel reduction for multi-channel audio, supporting both numpy arrays and torch tensors. Args: audio: Input audio data. Can be: - 1D array/tensor: (time,) - already mono - 2D array/tensor: (channels, time) - standard format from torchaudio - 2D array/tensor: (time, channels) - format from soundfile (will be auto-detected and transposed if time > channels) spec: AudioSpec defining the target format. Returns: Normalized audio in the same type as input (numpy or torch). For mono output (target_channels=1), returns 1D array/tensor. Raises: ValueError: If audio has unsupported dimensions or channel expansion is requested (e.g., mono to stereo). rzCannot expand mono audio to z channelszUnsupported audio shape: z. Expected 1D or 2D.rzCannot expand z channels to )axis)dimzUnknown reduction method: N)r%ndimr ValueErrorshape isinstancenpndarrayTrr rr rrrvaluesrr)r/r0 num_channelsis_numpyresults rnormalize_audior@Hs?2  #  zQ  1 $ $LW8LWWWXXX zQVU[VVVWWW {1~ A&&%eRZ88Eeg;q>Lt+++ d*** N\ N N8L N N   %,,H q   !%5%: : :/7NRWU++++UZZAZ=N=NFF  #'7'= = =1XFF  #'7'; ; ;.6SRVE****EII!I !T  ;) # #)w$. [G # #'w$. ?dk??? r)Nr) rrrrfloatrrPnptNDArrayr9floatingrErrrrMrMs66#'.74<*+{2;'  R[ ! rrM)! dataclassesrenumrtypingrnumpyr9 numpy.typingrTtorchvllm.utils.import_utilsrr ImportError scipy.signalr rIplaceholder_attrr.r rrMONO_AUDIO_SPECPASSTHROUGH_AUDIO_SPECrUrVTensorr@rSrGrKrMrrrrds"!!!!! 555555+NNNN+++ **GGG+I'''''''III$$W-->>xHHLLLI sD          >)AAQAVWWW"4888K- ;r{ #el 2K- K- [ ,K-K-K-K-fI ;r{ #II I  [ IIII  ;r{ #      !!!!!!!!!!s -AAA #A10A1