o xi@sfddlmZddlZddlmZddlmZddlZddlZdd d Z dddZ eGdddZ dS)) annotationsN) dataclass)Pathlatent torch.Tensor patch_sizeintreturncCsR|dkr|S|j\}}}|||}|ddd|f}||||||}|S)ze Convert latent from (B, T, D) -> (B, T//patch, D*patch). Extra tail tokens are dropped. N)shapereshape)rrbszseq_lendimusabler-/home/ubuntu/Irodori-TTS/irodori_tts/codec.pypatchify_latent s  rpatched latent_dimcCs*|dkr|S||jd|jd||S)zE Convert latent from (B, T_p, D*patch) -> (B, T_p*patch, D). r r)r r )rrrrrrunpatchify_latentsrc@seZdZUded<ded<ded<ded<ded <d ed <d ed <e     d&d'ddZedddd(ddZed)d d!Z d*d$d%Z dS)+ DACVAECodecztorch.nn.Modulemodelr sample_raterz torch.devicedevicez torch.dtypedtypeboolenable_watermark float | Nonewatermark_alphafacebook/dacvae-watermarkedcudaNFrepo_idstrtorch.dtype | Noner c Cszddlm}Wn'ty/ttjdd}|r'tj dt |ddlm}Ynw| | |}|durD|j|d}d} d} t|dd} | durt| drt| j} |duret|} n|rj| } nd } t| | _t| j} | d k} | st| d rd| fddd}|| _t|j}tjddd||d}t ||}Wdn1swY||t|jt|jdt||| | dS)Nr)DACVAEdacvaerFdecoderalphagwm_modelxrmessagetorch.Tensor | Noner cSs~|jj|S)N)r+ encoder_blockforward_no_conv)r,r-_decoderrrr_watermark_passthroughRsz0DACVAECodec.load.._watermark_passthroughr irr)rrrrrrr)r,rr-r.r r)r'r% ImportErrorr__file__resolveparentsexistssyspathinsertr#loadevaltogetattrhasattrfloatr* watermarknext parametersrtorchzerosinference_modeencoderrr r)clsr"rrrrr% local_reporconfigured_watermark_alphaconfigured_enable_watermarkr) default_alpha target_alphar2 model_dtypedummyzrrrr<,sV            zDACVAECodec.load) normalize_db ensure_maxwaveformrrRrSc Csz|jdkr |d}|jdkrtdt|j|jddkr'|jddd}||jkr5tj |||j}|j t j d}|d urzBtj ||j}|jdkrS|d}t||jd d d }t t jd |j|jd|dddd}t |}t ||||}Wn tyYnw|r|jdddjddd} | jdd} || }|j |j|jd}|j|} | ddS)a Input: waveform: (B, C, T) or (C, T) normalize_db: Optional target loudness (LUFS-like dB) applied before encode ensure_max: If True, scale down only when abs peak exceeds 1.0 Output: latent: (B, T_latent, D_latent) r&rz$Expected waveform ndim=3, got shape=r T)rkeepdimr(NgTgT@)minmaxg$@r3g4@g?)rW)ndim unsqueeze ValueErrortupler meanr torchaudio functionalresampler>rEfloat32loudnessrAclamppowtensorrrviewisfinitewhere ExceptionabsamaxrrH transpose contiguous) selfrTrrRrSrcgain_dbgain finite_maskpeak safe_peakencodedrrrencode_waveformksB           zDACVAECodec.encode_waveformrcCsH|jdkrtdt|j|ddj|j|jd}|j |S)ze Input: latent: (B, T, D) Output: audio: (B, 1, samples) rUz"Expected latent ndim=3, got shape=r r&r() rZr\r]r rmrnr>rrrdecode)rorrQrrr decode_latents  zDACVAECodec.decode_latentr: str | PathcCsz tt|\}}Wn+ty6ddl}|jt|dd\}}t|}|jdkr1| d}n|j }Ynw| d}| || S)Nrrbr(r ) r_r<r# RuntimeError soundfilereadrE from_numpyrZr[Trvcpu)ror:wavsrsfdatarrr encode_files     zDACVAECodec.encode_file)r r!NFN) r"r#rr#rr$rrrrr r) rTrrrrRrrSrr r)rrr r)r:ryr r) __name__ __module__ __qualname____annotations__ classmethodr<rErGrvrxrrrrrr"s, >4  r)rrrrr r)rrrrrrr r) __future__rr9 dataclassesrpathlibrrEr_rrrrrrrs