љic SSKrSSKrSSKJrJr SSKrSSKJr /SQr"SS5r "SS5r S5S \ S \ S \R4S jjrS5S \ S\RS \ 4Sjjr\R "S54S\ 4SjjrSr\R&"5SS S\"S5SSSSSSSSSS4S\RS\S \ S \ S!\S"\ S#\ S$\S%\ S&\S'\\/S4S(\S)\ S*\ S+\4S,jj5r"S-S.5rS6S/\\S0\RS1\S \ S \R4 S2jjrS6S/\\S0\RS1\S \ S \R4 S3jjrS/\\S \ S \\4S4jrg)7N)CallableList)version)ruendeescR\rSrSrS SjrS\4SjrS SjrS\4SjrS\4Sjr Sr g ) OnnxWrapper cZSSKqSSKnUR5nSUlSUlU(a,SUR 5;aURUS/US9UlOURXS9UlUR5 SU;a[R"S5 S/Ul gS S/Ul g) NrCPUExecutionProvider) providers sess_options)r16kz,This model support only 16000 sampling rate!>@) numpynp onnxruntimeSessionOptionsinter_op_num_threadsintra_op_num_threadsget_available_providersInferenceSessionsession reset_stateswarningswarn sample_rates)selfpathforce_onnx_cpuroptss N/home/james-whalen/.local/lib/python3.13/site-packages/silero_vad/utils_vad.py__init__OnnxWrapper.__init__ s))+$%!$%! 4 8[8[8]]&77I_H`os7tDL&777PDL  D= MMH I!&D !%u D srcUR5S:XaURS5nUR5S:a[SUR535eUS:waUS-S:XaUS-nUSS2SSU24nSnX R;a[SURS35eX!RS- S:a [S 5eX4$) Nrrz*Too many dimensions for input audio chunk rzSupported sampling rates: z (or multiply of 16000)g@?@zInput audio chunk is too short)dim unsqueeze ValueErrorr!shape)r"xr*steps r&_validate_inputOnnxWrapper._validate_input!s 557a< AA 557Q;I!%%'ST T ;BJ!O;D!FdF( AB && &9$:K:K9LLcde e  ?U "=> >u r)c[R"SUS45R5Ul[R"S5UlSUlSUlg)Nr,r)torchzerosfloat_state_context_last_sr_last_batch_size)r" batch_sizes r&rOnnxWrapper.reset_states3s@kk1j#"67==?  A  !r)cURX5upUS:XaSOSnURSU:wa[SURSS35eURSnUS:XaSOS nUR(dUR U5 UR (a!UR U:waUR U5 UR(a!URU:waUR U5 [ UR5(d[R"XE5Ul[R"URU/S S 9nUS ;azUR5URR5[RUS S9S.nURR!SU5nUup[R""U 5Ul O [5eUSU*S24UlX lX@l[R""U5nU$)NrzProvided number of samples is z< (Supported values: 256 for 8000 sample rate, 512 for 16000)r@ rr-rrint64)dtype)inputstater*.)r3r0r/r=rr<lenr;r7r8catrr:rarrayrrun from_numpy) r"r1r* num_samplesr> context_size ort_inputsort_outsoutrKs r&__call__OnnxWrapper.__call__9s$$Q+5[cc 772;+ %=aggbk]KGHI IWWQZ 5[rb $$   j ) MM  3   j )  ! !(=(=(K   j )4==!!!KK ADM IIt}}a(a 0  #$779t{{7H7H7JRTRZRZ[]elRZRmnJ||''j9H!JC**51DK, # }~-.  *s# r)c/nURX5upUR5 US:XaSOSnURSU-(a@XARSU-- n[RR R USU4SSS9n[SURSU5H2nUSS2XfU-24nURXr5nURU5 M4 [R"USS 9n U R5$) NrrArBrrconstantg)valuerF) r3rr0r7nn functionalpadrangerVappendrMcpu) r"r1r*outsrQpad_numi wavs_batch out_chunkstackeds r& audio_forwardOnnxWrapper.audio_forward^s$$Q+ 5[cc 771: #!WWQZ+%=>G##''Aw<3'OAq!''!*k2A1a+ o-.J j5I KK "3 ))Da({{}r))r;r=r<r:r!rN)F)r) __name__ __module__ __qualname____firstlineno__r'intr3rrVrg__static_attributes__r)r&r r s1.*S$" #c#J3r)r c<\rSrSrSrS\R 4SjrSrg) ValidatorqcZURS5(aSOSUl[RR US5 UR(aMSSKnU(a+SUR 5;aURSS/S9UlgURS5Ulg[SS9Ulg) Nz.onnxTFz inf.modelrr)r) model_path) endswithonnxr7hubdownload_url_to_filerrrmodelinit_jit_model)r"urlr$rs r&r'Validator.__init__rsLL11Du  &&sK8 99 "8K<_<_/ , ood+GC  <&t,<<>ll(( ")**@*@)AAefBC s*0C:2D:DF4E+E==FtensorcUR5R5nURS:XaURS5n[R "[ R5n[ R"XUSS9 g![ak U[R "S5:aOSSK J n U"USS9nURU5 g![a! [S [ RS 3S -5ef=fef=f) Nrr)bits_per_sampler) AudioEncoderr)rrz! requires torchcodec for saving. r)detachr`rr.rrrrsave Exceptiontorchcodec.encodersrto_filerr)r#rrrrencoders r& save_audiors ]]_ "F {{a!!!$ ]]:11 2FmRH   W]]5) ) <&v5A% ")**@*@)AAbcBC   s%A<<$C1!C+C,,C1r`rtc`[RRXS9nUR5 U$)N) map_location)r7jitreval)rtdevicerys r&rzrzs$ IINN:N ;E JJL Lr)c SSKnURSU0[[U55Vs/sHo3U-PM snS9R SSSS/S[U5U-/SSS S 9 gs snf) Nrprobs)index)rareag?secondszspeech probabilitytab20)figsizekindylimxlimxlabelylabelcolormap)pandas DataFramer^rLplot)rr2pdr1s r&make_visualizationrssLL'5!*/E *;<*;QD*;<>>Bd7At9As5zD7H3I!,! ?C?#-S - n[ U5n/n[ SUU 5HnUUUU -n[ U5U :aA[RRRUS[U [ U5- 545nU"UU5R!5nUR#U5 UU -nUU:aUnUU- S-nU (dMU "U5 M Sn/n0n U c[%US- S5n Sn!S=n"n#/n$['U5GHunnU U-n%UU:a/U!(a(U%U!- n&U&U:aU$R#U!U&45 Sn!U#U":aU%n#UU:aU(d SnU%U S'MWU(aU%U S- U:aU(aLU$(aE[%U$SS9un"n'U"U S'UR#U 5 0n U"U'-n#U#U"U%-:aU#U S'OSnS=n"=n#n!/n$O[U"(a/U"U S'UR#U 5 0n U#U":aSnOU#U S'S=n"=n#n!/n$O%U%U S'UR#U 5 0n S=n"=n#n!Sn/n$GMUU :dGM!U(dGM+U!(dU%n!U%U!- n(U(dU(U:aU!n"U(U:aGMQU!U S'U SU S- U:aUR#U 5 0n S=n"=n#n!Sn/n$GM U (a"UU S- U:aUU S'UR#U 5 ['U5GHunn)US:Xa[[%SU)SU- 55U)S'U[ U5S- :waUUS-SU)S- n*U*S U-:aHU)S==[U*S -5- ss'[[%SUUS-SU*S -- 55UUS-S'M[[)UU)SU-55U)S'[[%SUUS-SU- 55UUS-S'M[[)UU)SU-55U)S'GM U(aMUU- n+UHAn,[%[+U,SU- U 5S5U,S'[)[+U,SU- U 5U+5U,S'MC O)US:a#UHn,U,S==U-ss'U,S==U-ss'M U (a[-UX- 5 U$! [S5e=f)a This method is used for splitting long audios into speech chunks using silero VAD Parameters ---------- audio: torch.Tensor, one dimensional One dimensional float torch.Tensor, other types are casted to torch if possible model: preloaded .jit/.onnx silero VAD model threshold: float (default - 0.5) Speech threshold. Silero VAD outputs speech probabilities for each audio chunk, probabilities ABOVE this value are considered as SPEECH. It is better to tune this parameter for each dataset separately, but "lazy" 0.5 is pretty good for most datasets. sampling_rate: int (default - 16000) Currently silero VAD models support 8000 and 16000 (or multiply of 16000) sample rates min_speech_duration_ms: int (default - 250 milliseconds) Final speech chunks shorter min_speech_duration_ms are thrown out max_speech_duration_s: int (default - inf) Maximum duration of speech chunks in seconds Chunks longer than max_speech_duration_s will be split at the timestamp of the last silence that lasts more than 100ms (if any), to prevent aggressive cutting. Otherwise, they will be split aggressively just before max_speech_duration_s. min_silence_duration_ms: int (default - 100 milliseconds) In the end of each speech chunk wait for min_silence_duration_ms before separating it speech_pad_ms: int (default - 30 milliseconds) Final speech chunks are padded by speech_pad_ms each side return_seconds: bool (default - False) whether return timestamps in seconds (default - samples) time_resolution: bool (default - 1) time resolution of speech coordinates when requested as seconds visualize_probs: bool (default - False) whether draw prob hist or not progress_tracking_callback: Callable[[float], None] (default - None) callback function taking progress in percents as an argument neg_threshold: float (default = threshold - 0.15) Negative threshold (noise or exit threshold). If model's current state is SPEECH, values BELOW this value are considered as NON-SPEECH. min_silence_at_max_speech: int (default - 98ms) Minimum silence duration in ms which is used to avoid abrupt cuts when max_speech_duration_s is reached use_max_poss_sil_at_max_speech: bool (default - True) Whether to use the maximum possible silence at max_speech_duration_s or not. If not, the last silence is used. window_size_samples: int (default - 512 samples) !!! DEPRECATED, DOES NOTHING !!! Returns ---------- speeches: list of dicts list containing ends and beginnings of speech chunks (samples or seconds based on return_seconds) 2Audio cannot be casted to tensor. Cast it manuallyrrzRMore than one dimension in audio. Are you trying to process audio with 2 channels?rNz@Sampling rate is a multiply of 16000, casting to 16000 manually!rGzVCurrently silero VAD models support 8000 and 16000 (or multiply of 16000) sample ratesrArBr,rF333333?g{Gz?Tstartc US$)Nrro)r1s r&'get_speech_timestamps..qs1r))keyend)r7 is_tensorr TypeErrorrLr0r^rr/rr rr[r\r]rmitemr_max enumerateminroundr)-rryrrrrrrrrrrrrrrrcr2min_speech_samplesspeech_pad_samplesmax_speech_samplesmin_silence_samples!min_silence_samples_at_max_speechaudio_length_samples speech_probscurrent_start_samplechunk speech_probprogressprogress_percent triggeredspeechescurrent_speechtemp_endprev_end next_start possible_ends cur_samplesil_durdur sil_dur_nowspeechsilence_durationaudio_length_seconds speech_dicts- r&get_speech_timestampsr s\ ??5 ! ! RLL'E 5;;!s5;;'(AMM!$E) u{{ a qr ru-%"71"<% ff  XYM)qrr!.%!7#S &?$F&6=&>ATTWX[mWmm'ADH(5(QTX(X%u:L %a)=?R S*,@CV,VW u:+ +HH''++EAs;NQTUZQ[;[7\3]^EE=1668 K('*== * *+H$';;sB % % &'7 8!TIHNI,d3 HHzM#L1;(1,  9 $( 8+G::$$h%89HH$'  9 $iI&0N7 #  *~g'>>ASS-- #M~ F #(0u%/!#%^ : 55.8N7+ %I3444: " ,4N5)OON3%'N!H,$) 2<w/788H8zH$&M-7N5)OON3%'N788H8zH %I$&M - 'YY%$x/K1kDe6e#00(0u%"5)N7,CCGYYOON3!#3444:! " U2X/.2IIM__ 4u'x( 6 6!#a;M)M"NOF7O H ! !'!}W5u E !&8"88u %5%:!;; ),SHQqSM'4JM]abMb4b-c)d1 g& #C(`$a bu ),SHQqSM'4JM_4_-`)a1 g&$8&-J\:\ ]^F5M)3mC#K#&u[-AM-QSb'cef#gK !$U;u+= +M%_au!vK $ #K  D (  $ & $<)<)LM OG RPQ Qs U>> V c z\rSrSrS S\S\S\S\4SjjrSr\R"5S S\4S jj5r S r g ) VADIteratorirrrrcXlX lX0lUS;a [S5eX4-S- UlX5-S- UlUR 5 g)a Class for stream imitation Parameters ---------- model: preloaded .jit/.onnx silero VAD model threshold: float (default - 0.5) Speech threshold. Silero VAD outputs speech probabilities for each audio chunk, probabilities ABOVE this value are considered as SPEECH. It is better to tune this parameter for each dataset separately, but "lazy" 0.5 is pretty good for most datasets. sampling_rate: int (default - 16000) Currently silero VAD models support 8000 and 16000 sample rates min_silence_duration_ms: int (default - 100 milliseconds) In the end of each speech chunk wait for min_silence_duration_ms before separating it speech_pad_ms: int (default - 30 milliseconds) Final speech chunks are padded by speech_pad_ms each side rGzDVADIterator does not support sampling rates other than [8000, 16000]rN)ryrrr/rrr)r"ryrrrrs r&r'VADIterator.__init__sT: "*  -cd d#0#JT#Q "/"?$"F r)cbURR5 SUlSUlSUlg)NFr)ryrrrcurrent_sample)r"s r&rVADIterator.reset_statess( ! r)rc([R"U5(d[R"U5nUR 5S:Xa[ US5O [ U5nU=R U- slURXR5R5nXPR:aUR(aSUl XPR:akUR(dZSUl [SUR UR- U- 5nSU(d [U50$[!X`R- U50$XPRS- :aUR(aUR(dUR Ul UR UR- UR":agURUR-U- nSUl SUl S U(d [U50$[!XpR- U50$g! [S5e=f) a x: torch.Tensor audio chunk (see examples in repo) return_seconds: bool (default - False) whether return timestamps in seconds (default - samples) time_resolution: int (default - 1) time resolution of speech coordinates when requested as seconds rr,rTrrNFr)r7rrrr-rLrryrrrrrrrrmrr)r"r1rrrr speech_start speech_ends r&rVVADIterator.__call__sq!! VLLO,-557a >> )t}}DM >> )4>>!DNq$"5"58O8O"ORe"efLnS.E E%P\_q_qPqtCKDE E ..4/ /T^^== $ 3 3 ""T]]2T5M5MM!]]T-D-DDGZZ ! !&ns:CC%PZ]o]oPorAKBCC7 V TUUs H H)rrryrrrrrN)rrrr)Fr) rirjrkrlr9rmr'rr7rrVrnror)r&r r sb%(&+03&( &!&!$&+. & !$ &P  ]]_***r)r tssrrcU(aU(d [S5e[5nU(a [X5OUnUHnURXSUS5 M [R "U5$)aCollect audio chunks from a longer audio clip This method extracts audio chunks from an audio clip, using a list of provided coordinates, and concatenates them together. Coordinates can be passed either as sample numbers or in seconds, in which case the audio sampling rate is also needed. Parameters ---------- tss: List[dict] Coordinate list of the clips to collect from the audio. wav: torch.Tensor, one dimensional One dimensional float torch.Tensor, containing the audio to clip. seconds: bool (default - False) Whether input coordinates are passed as seconds or samples. sampling_rate: int (default - None) Input audio sampling rate. Required if seconds is True. Returns ------- torch.Tensor, one dimensional One dimensional float torch.Tensor of the concatenated clipped audio chunks. Raises ------ ValueError Raised if sampling_rate is not provided when seconds is True. 3sampling_rate must be provided when seconds is Truerrr/list_seconds_to_samples_tssr_r7rM)rrrrchunks_tssrcs r&collect_chunksr(s`D}NOO VF:A "3 6sD  cG*QuX./ 99V r)cU(aU(d [S5e[5nSnU(a [X5OUnUHnURXUS5 USnM URXS5 [R "U5$)aDrop audio chunks from a longer audio clip This method extracts audio chunks from an audio clip, using a list of provided coordinates, and drops them. Coordinates can be passed either as sample numbers or in seconds, in which case the audio sampling rate is also needed. Parameters ---------- tss: List[dict] Coordinate list of the clips to drop from from the audio. wav: torch.Tensor, one dimensional One dimensional float torch.Tensor, containing the audio to clip. seconds: bool (default - False) Whether input coordinates are passed as seconds or samples. sampling_rate: int (default - None) Input audio sampling rate. Required if seconds is True. Returns ------- torch.Tensor, one dimensional One dimensional float torch.Tensor of the input audio minus the dropped chunks. Raises ------ ValueError Raised if sampling_rate is not provided when seconds is True. rrrrNr)rrrrr cur_startrrcs r& drop_chunksr VszD}NOO VFI:A "3 6sD  saj13eH  MM#j/" 99V r)ctUVs/sH&n[US5U-[US5U-S.PM( sn$s snf)zDConvert coordinates expressed in seconds to sample coordinates. rr)rr)r)rrcrds r&rrsQ  #s7|$}4SZ =0   s-5)r)FN)r7rtypingrrr packagingr languagesr rqrrmrrrrrzrrr9boolr r dictrr rror)r&r(sA ! $ ddN2S@S%,,s2 ,,u-s#.1/48;9>u9