o Si @sUddlZddlZddlZddlZddlZddlZddlmZddlmZddl m Z m Z ddl m Z ddlmZmZmZddlmZmZmZmZmZmZmZddlZddlZddlmZmZm Z m!Z!dd l"m#Z#dd l$m%Z%m&Z&m'Z'm(Z(e(a)e*e+d <da,ed e+d <dZ-dee.fddZ/eded e.fdedfddZ0d|ddZ1ded e.fdd fddZ2eddd|ddZ3de*ddfdd Z4de*fd!d"Z5eZ6Gd#d d Z7Gd$d%d%e7Z8Gd&d'd'e7Z9Gd(d)d)e7Z:Gd*d+d+e7Z;Gd,d-d-e7ZGd2d3d3e7Z?Gd4d5d5eZ@eddde*fd6d7ZAeddde*fd8d9ZBeddde*fd:d;ZCdZDd?ee e.e fde@fd@dAZEd?ee e.e fde@fdBdCZF d}dDe%dEe&dFee&deejGeHffdGdHZI  d~dDee%e fdEe&dFee&dIeeHdeejGeHff dJdKZJ d}d?ee e.e fdEe&dFee&deejGeHffdLdMZK d}dDe%dEe&dFee&deejGeHffdNdOZLdPe%de@fdQdRZMedddSdTZNdUdejOfdVee%e6fdEe&dFe&fdWdXZPdYejOfdZd[ZQ ddPe%d\eeHde@fd]d^ZR U  ddPe%dEe&dFee&d\eeHdeejGeHff d_d`ZS U  ddPe%dEe&dFee&d\eeHdeejGeHff dadbZT U  ddPe%dEe&dFee&d\eeHdeejGeHff dcddZUdeeVde.fdfdgZWdPe%de@fdhdiZXdPe%de@fdjdkZY ddle%dEe&dFee&deejGeHffdmdnZZ  ddoee.e e fdpeej[ejGfdqeHdree.dsee.ddf dtduZ\ U  ddDee%e6fdEe&dFee&d\eeHdeejGeHff dvdwZ]  xddPee%e fd\eeHdye*de@fdzd{Z^dS)N)contextmanager) lru_cache)BytesIOIOBase)Path)PIPECalledProcessErrorrun)Any GeneratorList NamedTupleOptionalTupleUnion)AudioLoadingErrorAudioSavingError VideoInfo verbose_audio_loading_exceptions)Resample)PathlikeSecondscompute_num_samplesis_torchaudio_available_FFMPEG_TORCHAUDIO_INFO_ENABLED AudioBackendCURRENT_AUDIO_BACKEND)z.aviz.mov.mp4.m4az.wmvz.mkvz.webmz.flvreturncCsdgttjS)zR Return a list of names of available audio backends, including "default". default)sortedrKNOWN_BACKENDSkeysr$r$H/home/ubuntu/.local/lib/python3.10/site-packages/lhotse/audio/backend.pyavailable_audio_backends)sr&backend)rNNccs"t}t|}|Vt|dS)a Context manager that sets Lhotse's audio backend to the specified value and restores the previous audio backend at the end of its scope. Example:: >>> with audio_backend("LibsndfileBackend"): ... some_audio_loading_fn() N)get_current_audio_backendset_current_audio_backend)r'previousbr$r$r% audio_backend0s   r,cCs8tdurtStjd}|durt|tStdtS)zI Return the audio backend currently set by the user, or default. NLHOTSE_AUDIO_BACKENDr )rosenvirongetr)) maybe_backendr$r$r%r(Cs r(cCsb|dkr t}|atSt|trt|}|atSt|tr!|}t|ts-Jd||atS)a5 Force Lhotse to use a specific audio backend to read every audio file, overriding the default behaviour of educated guessing + trial-and-error. Example forcing Lhotse to use ``audioread`` library for every audio loading operation:: >>> set_current_audio_backend(AudioreadBackend()) r z#Expected str or AudioBackend, got: )get_default_audio_backend isinstancestrrnewtyperr'r$r$r%r)Xs"    r))maxsizecCsFg}tjdddkr|t|ttttt g7}t |S)a Return a backend that can be used to read all audio formats supported by Lhotse. It first looks for special cases that need very specific handling (such as: opus, sphere/shorten, in-memory buffers) and tries to match them against relevant audio backends. Then, it tries to use several audio loading libraries (torchaudio, soundfile, audioread). In case the first fails, it tries the next one, and so on. LHOTSE_LEGACY_OPUS_LOADING01) r.r/r0appendFfmpegSubprocessOpusBackendSph2pipeSubprocessBackendLibsndfileBackendTorchaudioFFMPEGBackendTorchaudioDefaultBackendAudioreadBackendCompositeAudioBackendbackendsr$r$r%r2ps   r2enabledcCs(|tkrtdtd|d|adS)a  Override Lhotse's global setting for whether to use ffmpeg-torchaudio to compute the duration of audio files. If disabled, we fall back to using a different backend such as sox_io or soundfile. .. note:: See this issue for more details: https://github.com/lhotse-speech/lhotse/issues/1026 Example:: >>> import lhotse >>> lhotse.set_ffmpeg_torchaudio_info_enabled(False) # don't use ffmpeg-torchaudio :param enabled: Whether to use torchaudio to compute audio file duration. zThe user overrided the global setting for whether to use ffmpeg-torchaudio to compute the duration of audio files. Old setting: z. New setting: .N)rlogginginfo)rGr$r$r%"set_ffmpeg_torchaudio_info_enabledsrKcCstS)a Return FFMPEG_TORCHAUDIO_INFO_ENABLED, which is Lhotse's global setting for whether to use ffmpeg-torchaudio to compute the duration of audio files. Example:: >>> import lhotse >>> lhotse.get_ffmpeg_torchaudio_info_enabled() )rr$r$r$r%"get_ffmpeg_torchaudio_info_enableds rLc s2eZdZdZiZfddZededdfddZ  d!d e e e fd e d e e d e edeejeff ddZ d"d e e e fd e efddZd e e e fdefddZd e e e fdefddZdefddZdefddZ d#de eeefde ejejfdede ede edd f dd ZZS)$ra Internal Lhotse abstraction. An AudioBackend defines three methods: one for reading audio, and two filters that help determine if it should be used. ``handles_special_case`` means this backend should be exclusively used for a given type of input path/file. ``is_applicable`` means this backend most likely can be used for a given type of input path/file, but it may also fail. Its purpose is more to filter out formats that definitely are not supported. c s.|jtjvr |tj|j<tjdi|dS)Nr$)__name__rr"super__init_subclass__)clskwargs __class__r$r%rOs  zAudioBackend.__init_subclass__namercCs$||jvr td||j|S)NzUnknown audio backend name: )r" RuntimeError)rPrTr$r$r%r5s  zAudioBackend.newN path_or_fdoffsetdurationforce_opus_sampling_ratecCtNNotImplementedErrorselfrWrXrYrZr$r$r% read_audioszAudioBackend.read_audiocCr[r\r]r`rWrZr$r$r%rJszAudioBackend.infocCdSNFr$r`rWr$r$r%handles_special_casez!AudioBackend.handles_special_casecCrcNTr$rer$r$r% is_applicablergzAudioBackend.is_applicablecCrcrdr$r`r$r$r% supports_savergzAudioBackend.supports_savecCrcrdr$rjr$r$r% supports_inforgzAudioBackend.supports_infodestsrc sampling_rateformatencodingcCr[r\r])r`rmrnrorprqr$r$r% save_audioszAudioBackend.save_audiorVNNr\NN)rM __module__ __qualname____doc__r"rO classmethodr4r5rr FileObjectrrintrnpndarrayrarJboolrfrirkrlrrtorchTensorrr __classcell__r$r$rRr%rsZ        c @seZdZ   ddeeefdedeedeefddZ deeefd e fd d Z deeefd e fd d Z d e fddZ ddeeefdeefddZdS)r>rVNrWrXrYrZcCs2t|ttfsJdt|dt||||dS)Nzs,   r>c @eZdZ   ddeeefdedeedeede e j eff dd Z deeefde fd d Zdeeefde fd d Zde fddZ ddeeefdeefddZdS)r?rVNrWrXrYrZrcCs0t|ttfsJdt|dt|||dS)Nz>Cannot use an sph2pipe subprocess to read from path of type: 'r)sph_pathrXrY)r3r4rr6read_sphr_r$r$r%ra"sz$Sph2pipeSubprocessBackend.read_audiocCr)Nz.sphrrer$r$r%rf2rz.Sph2pipeSubprocessBackend.handles_special_casecCrr\rrer$r$r%ri7rz'Sph2pipeSubprocessBackend.is_applicablecCrcrhr$rjr$r$r%rl:rgz'Sph2pipeSubprocessBackend.supports_infocCt|Sr\)sph_inforbr$r$r%rJ=zSph2pipeSubprocessBackend.inforsr\rMrurvrrryrrrzrr{r|rar}rfrirlrJr$r$r$r%r?!s0    r?c @r)FfmpegTorchaudioStreamerBackendrVNrWrXrYrZrcCt|||dS)N)path_or_fileobjrXrY)torchaudio_ffmpeg_loadr_r$r$r%raF z*FfmpegTorchaudioStreamerBackend.read_audiocCto to t|tSr\rtorchaudio_supports_ffmpegr3rrer$r$r%rfSs z4FfmpegTorchaudioStreamerBackend.handles_special_casecCrr\rrer$r$r%riZs z-FfmpegTorchaudioStreamerBackend.is_applicablecCrcrhr$rjr$r$r%rlcrgz-FfmpegTorchaudioStreamerBackend.supports_infocCrr\torchaudio_inforbr$r$r%rJfrz$FfmpegTorchaudioStreamerBackend.inforsr\rr$r$r$r%rEs0      rc @seZdZ   ddeeefdedeedeede e j eff dd Z deeefde fd d Zde fd d Z  ddeeeefdeeje j fdedeedeeddf ddZde fddZ ddeeefdeefddZdS)rBrVNrWrXrYrZrcCrNrWrXrY)torchaudio_loadr_r$r$r%raorz#TorchaudioDefaultBackend.read_audiocCstSr\)rrer$r$r%ri|sz&TorchaudioDefaultBackend.is_applicablecCrcrhr$rjr$r$r%rkrgz&TorchaudioDefaultBackend.supports_savermrnrorprqcCsddl}ts Jdt|}t|trt|}|dkp)t|t o)| d}|r:|j j j |||ddddS|j |||||ddS)Nrz8We don't support saving audio with torchaudio pre v0.9.0flacz.flac) sample_raterpbits_per_sample)rrprq) torchaudio$torchaudio_soundfile_supports_formatr~ as_tensorr3rr4rrrr'soundfile_backendsave)r`rmrnrorprqr saving_flacr$r$r%rrs,      z#TorchaudioDefaultBackend.save_audiocCrcrhr$rjr$r$r%rlrgz&TorchaudioDefaultBackend.supports_infocCrr\rrbr$r$r%rJrzTorchaudioDefaultBackend.inforsrtr\)rMrurvrrryrrrzrr{r|rar}rirkr4rrr~rrrrlrJr$r$r$r%rBnsN      $ rBc @eZdZdZ   ddeeefdedeedee de e j e ff d d Z deeefdefd d Zdeeefdefd dZdefddZ  ddeeeefdeeje j fde deedeeddf ddZdefddZ ddeeefdee fddZdS)rAz A new FFMPEG backend available in torchaudio 2.0. It should be free from many issues of soundfile and sox_io backends. rVNrWrXrYrZrcCst||||dS)N)rWrXrY resample_rate)torchaudio_2_ffmpeg_loadr_r$r$r%ras z"TorchaudioFFMPEGBackend.read_audiocs(tt }| otfddtDS)Nc3|] }t|VqdSr\r4r.0extrWr$r%  z?TorchaudioFFMPEGBackend.handles_special_case..)r3ranySUPPORTED_VIDEO_EXTENSIONS)r`rW is_fileobjr$rr%rfs z,TorchaudioFFMPEGBackend.handles_special_casecCs totS)z FFMPEG backend requires at least Torchaudio 2.0. For version == 2.0.x, we also need env var TORCHAUDIO_USE_BACKEND_DISPATCHER=1 For version >= 2.1.x, this will already be the default. )r#torchaudio_ffmpeg_backend_availablerer$r$r%ris z%TorchaudioFFMPEGBackend.is_applicablecCrcrhr$rjr$r$r%rkrgz%TorchaudioFFMPEGBackend.supports_savermrnrorprqcCsHddl}t|st|}t|trt|}|j|||||dddS)Nrffmpeg)rrprqr')rr~ is_tensorrr3rr4r)r`rmrnrorprqrr$r$r%rrs    z"TorchaudioFFMPEGBackend.save_audiocCrcrhr$rjr$r$r%rlrgz%TorchaudioFFMPEGBackend.supports_infocCrr\)torchaudio_ffmpeg_streamer_inforbr$r$r%rJrzTorchaudioFFMPEGBackend.inforsrtr\rMrurvrwrrryrrrzrr{r|rar}rfrirkr4rrr~rrrrlrJr$r$r$r%rAsR      rAc @r)r@a A backend that uses PySoundFile. .. note:: PySoundFile has issues on MacOS because of the way its CFFI bindings are implemented. For now, we disable it on this platform. See: https://github.com/bastibe/python-soundfile/issues/331 rVNrWrXrYrZrcCrr)soundfile_loadr_r$r$r%rarzLibsndfileBackend.read_audiocCs8t|tr ts dSt|ttfrt|drdSdS)NTrF)r3rrrr4rrer$r$r%rfsz&LibsndfileBackend.handles_special_casecs,tttfrtfdddDrdSdS)Nc3rr\rrrr$r%rrz2LibsndfileBackend.is_applicable..)rrz.m4bFT)r3rr4rrer$rr%ris zLibsndfileBackend.is_applicablecCrcrhr$rjr$r$r%rk rgzLibsndfileBackend.supports_savermrnrorprqcCsVddl}t|r |}t|trt|}|dkrd}d}|j||j|||ddS)Nropusogg)filedata sampleraterpsubtype) soundfiler~rnumpyr3rr4writeT)r`rmrnrorprqsfr$r$r%rr#s   zLibsndfileBackend.save_audiocCrcrhr$rjr$r$r%rl<rgzLibsndfileBackend.supports_infocCrr\)soundfile_inforbr$r$r%rJ?rzLibsndfileBackend.inforsrtr\rr$r$r$r%r@sR         r@c @sveZdZ   ddeeefdedeedeede e j eff dd Z d d Z ddeeefdeefd d ZdS)rCrVNrWrXrYrZrcCr)N) path_or_filerXrY)audioread_loadr_r$r$r%raHrzAudioreadBackend.read_audiocCrcrhr$rjr$r$r%rlUrgzAudioreadBackend.supports_infocCrr\)audioread_inforbr$r$r%rJXrzAudioreadBackend.inforsr\)rMrurvrrryrrrzrr{r|rarlrJr$r$r$r%rCGs,     rCc @seZdZdZdeefddZ   ddeee fde d e e d e e d e eje ff d d Zd efddZ  ddeeeefdeejejfde de ede ed df ddZd efddZ ddeee fd e e fddZdS)rDa Combines multiple audio backends. It will try each out sequentially, and back off to the next one in the list if the current one fails. It uses the special filter methods to prioritize special case backends, and skip backends that are not applicable. rFcCs ||_dSr\rE)r`rFr$r$r%__init__hrzCompositeAudioBackend.__init__rVNrWrXrYrZrc Csg}|jD] }||r||qt|dks Jd|dt|dkrPz |dj||||dWStyO}ztd|dt|d t|d}~wwg}|jD]R}| |rz |j||||dWSty}z2d t|d t|d } t r|| t n|| t|d t|WYd}~qUd}~wwqU|std |dd} t rdnd} td|d| | || )NaCompositeAudioBackend has more than one sub-backend that handles a given special case for input 'rr8rrWrXrYrZzReading audio from '' failed. Details: : Exception # (): (No applicable backend found for input: ' Y Set LHOTSE_AUDIO_LOADING_EXCEPTION_VERBOSE=1 environment variable for full stack traces.' failed. Details:)rFrfr=lenra Exceptionrr6r4rir traceback format_excjoin) r`rWrXrYrZ candidatesr+e exceptionsmsgNL maybe_infor$r$r%raksl         z CompositeAudioBackend.read_audiocCstdd|jDS)Ncss|]}|VqdSr\rkrr+r$r$r%rsz6CompositeAudioBackend.supports_save..)rrFrjr$r$r%rksz#CompositeAudioBackend.supports_savermrnrorprqc Csdd|jD}g}|D]S}||sq z |j|||||dWSty_} z2dt|dt|d} trE|| t n|| t| dt | WYd} ~ q d} ~ ww|sft dd } trmd nd } t d | | || ) NcSg|]}|r|qSr$rrr$r$r% z4CompositeAudioBackend.save_audio..rmrnrorprqrrrrz-No applicable backend found for saving audio.rrrzSaving audio failed. Details:) rFrirrrrr6rr=rrr4rr) r`rmrnrorprqrrr+rrrrr$r$r%rrs<   z CompositeAudioBackend.save_audiocCrcrhr$rjr$r$r%rlrgz#CompositeAudioBackend.supports_infoc Csdd|jD}g}|D] }||r||q t|dks'Jd|dt|dkrTz |dj|dWStyS}ztd |d t|d t|d}~wwg}|D]O}| |rz |j|dWSty}z2d t|d t|d}t r||t n||t|d t|WYd}~qXd}~wwqX|std|dd} t rdnd} td |d| | || )NcSrr$)rlrr$r$r%rrz.CompositeAudioBackend.info..rrrr8rrz Fetching info about audio from 'rrrrrrrrrr)rFrfr=rrJrrr6r4rirrrr) r`rWrZrFrr+rrrrrr$r$r%rJsZ       zCompositeAudioBackend.inforsrtr\)rMrurvrwr rrrrryrrrzrr{r|rar}rkr4rrr~rrrrlrJr$r$r$r%rD`sP   >  * rDc@s>eZdZUeed<eed<eed<eed<dZeeed<dS)LibsndfileCompatibleAudioInfochannelsframesrrYNvideo) rMrurvrz__annotations__floatrrrr$r$r$r%r s rcCstsdStdS)zu Returns ``True`` when torchaudio version is at least 0.12.0, which has support for FFMPEG streamer API. Fz0.12.0)rcheck_torchaudio_version_gtr$r$r$r%rsrcCtotdS)zr Returns ``True`` when torchaudio.load supports "ffmpeg" backend. This requires either version 2.1.x+ z2.1.0rrr$r$r$r%r!rcCr)z Returns ``True`` when torchaudio version is at least 0.9.0, which has support for ``format`` keyword arg in ``torchaudio.save()``. z0.9.0rr$r$r$r%r*rrversioncCs4tsdSddl}ddlm}||j||kS)NFr)r)rr packagingrparse __version__)rr_versionr$r$r%r3s  rrc sLddl}tr)d|vrdnd}|j|d}t|j|jt|j|j|jdSt t t fo:t fdddD}t t }|sD|rtrdd lm}||rTt nd }|jd ksaJd ||j}|jt|jd d}|D] \} || jd7}qvt|j|t|j||jdS|}t|j|jt|j|j|jdS) Return an audio info data structure that's a compatible subset of ``pysoundfile.info()`` that we need to create a ``Recording`` manifest. rNrr7rrrrYc3rr\rrrr$r%rQrz"torchaudio_info..).mp3r StreamReaderrnr8jLhotse doesn't support files with more than one source stream yet (not to be confused with multi-channel).frames_per_chunk)rrlist_audio_backendsrJr num_channels num_framesrzrr3r4rrrr torchaudio.iornum_src_streamsget_src_stream_infodefault_audio_streamadd_basic_audio_streamstreamshape) rrr'rJis_mpegrrstreamer tot_sampleschunkr$rr%r=sT       rcsddlm}tt  otfdddtDs"t|}|j}i}i}t|D]!}| |}|j dkrB|||<q1|j dkrL|||<q1t d|t |d kscJd t |d t |d kssJd t |d i|rt |\\}} | j} | dkr|jt| j|d|D] \} | | jd7} q|dt| j| j| j| dd<|rt |\\} |jtj| dfdd} | }|dkr|D] \} || jd7}qۈjj|tj|jdn jddddjdtdiS)Nrrc3rr\rrrr$r%rs  z2torchaudio_ffmpeg_streamer_info..)rraudiozUnexpected media_type: rzNLhotse currently does not support more than one video stream in a file (found z).z_Lhotse currently does not support files with more than a single FFMPEG audio stream yet (found zP). Note that this is not the same as multi-channel which is generally supported.) stream_index)fpsheightwidthr )rrcsJsr"dd}|dur jj}t||jdkr jSdSjS)NrgMbP?r)r0r rabsrY) video_infoaudio_duration) audio_streamrrmetar$r%_try_read_num_sampless  z>torchaudio_ffmpeg_streamer_info.._try_read_num_samplesrr$) r rr3rrrr4r ranger  media_typerUrlistitemsr add_basic_video_streamround frame_raterr remove_streamrrrrrzrupdater rYr)rrr num_streams audio_streams video_streams stream_idxrJvideo_stream_idx video_stream tot_framesraudio_stream_idxr rr$)rrrrrr%rs           rrWrXrYc Csddl}t|tr t|}d}d}|dks|dur1t|}|dkr't||j}|dur1t||j}t|tr;|d|j |||d\}}| t |fS)Nr) frame_offsetr ) rr3rr4rrrrseekloadrrz) rWrXrYrr3r  audio_inforror$r$r%rs&      rrc Csddl}t|tr t|}d}d}|dks|dur4|j|dd}|dkr*t||j}|dur4t||j}t|tr>|d|j |||dd\}} | t | fS)Nrr2rr7)r3r r') rr3rr4rJrrrr4r5rrz) rWrXrYrrr3r r6rror$r$r%r s(      rcCsddl}ts tdt|trt|}|jj|d}|jdks$Jd| |j }t |j }|durO|j t||d||t|\}|dd}n|j |d||tjdd|Ddd }||fS) NrzcUsing FFMPEG streamer backend for reading is supported only with PyTorch 1.12+ and torchaudio 0.12+rr8rrcSsg|] \}|ddqS)rr8) transpose)rtr$r$r%rPsz*torchaudio_ffmpeg_load..)dim)rrrUr3rr4iorr r rrzrrrr4nextrr7r~catr)rrXrYrrrJrorr$r$r%r,s0        rcCsddl}t|ttfrdt|vrddlm}||}||1}|j}|dkr2| t |||durwYdS)rrNrEr8r) audioread audio_openr4_available_audioread_backendsrrrrrrzr)rrI input_filer num_samplesr$r$r%rps    $rcCs$ddl}|}td||S)z Reduces the overhead of ``audioread.audio_open()`` when called repeatedly by caching the results of scanning for FFMPEG etc. rNz%Using audioread. Available backends: )rIavailable_backendsrIrJ)rIrFr$r$r%rKsrKrVrcs\ddltfdd}g}|s}|j}|j}tt|||} |dur,tj} n | tt|||} d} |D]@} t| |d} | } | t | } | | krRq=| | krXn&| | krd| d| | } | | krn| krxnn| | | d} | | q=Wdn1swY|rt |}|dkr| d|fj }ntjd|d}|t|fS)zLoad an audio buffer using audioread. This loads one block at a time, and then concatenates the results. This function is based on librosa: https://github.com/librosa/librosa/blob/main/librosa/core/audio.py#L180 rNc3s0tttfrjtdVdSVdS)NrE)r3r4rrJrKr$rIrr$r% file_handles  z#audioread_load..file_handler@r8r2)rIrrrrzr{r&inf _buf_to_floatrr= concatenatereshaperempty)rrXrYr@rPyrL sr_native n_channelss_starts_endnframen_prevr$rOr%rsD    &  rrcCs8dtdd|d>}d|}|t|||S)a{Convert an integer buffer to floating point values. This is primarily useful when loading integer-valued wav data into numpy arrays. This function is based on librosa: https://github.com/librosa/librosa/blob/main/librosa/util/utils.py#L1312 Parameters ---------- x : np.ndarray [dtype=int] The integer-valued data buffer n_bytes : int [1, 2, 4] The number of bytes per sample in ``x`` dtype : numeric type The target output type (default: 32-bit float) Returns ------- x_float : np.ndarray [dtype=float] The input data buffer cast to floating point g?r8zr) rr3rr4rBr?rCrJrrrrrY)rrr?info_r$r$r%rs    rcCs6t|\}}t|jd|jdt||jd|dS)Nrr8r)rrrrz)rrgror$r$r%rs  rrc Cst|}d|d}|dur|t||d7}|d|7}ztt|ddttdj}WntyC}z |jdkr>td d}~wwd dl }| |$}|j t j d |j}} |jd krd|d d n|j}Wd|| fS1stwY|| fS)z Reads SPH files using sph2pipe in a shell subprocess. Unlike audioread, correctly supports offsets and durations for reading short chunks. :return: a tuple of audio samples and the sampling rate. zsph2pipe -f wav -p -t :N T)rmcheckrnrozXIt seems that 'sph2pipe' binary is not installed; did you run 'lhotse install-sph2pipe'?rrQr8r2)rr&rr rrnrrtrsrrDrEr{rFrrrUr) rrXrYrurvrrrGrror$r$r%rs2    rrmrnrorprqcCstj|||||dS)Nr)r(rrrr$r$r%rrsrrcCstj||||dS)Nr)r(rarr$r$r%ras raFforce_read_audiocCs8|rt|ttfs Jdtj|dStj||dS)Nz:force_read_audio=True does not work with file-like objectsr)rWrZ)r3r4rrCrJr()rrZrr$r$r%rJsrJ)rr)rN)rNNr\rs)rVNrtrd)_ contextlibrIr.r{sysrr functoolsrr:rrpathlibr subprocessrrr typingr r r r rrrrr{r~lhotse.audio.utilsrrrrlhotse.augmentationr lhotse.utilsrrrrrr}rrrr4r&r,r(r)r2rKrLryrr>r?rrBrAr@rCrDrrrrrrrr|rzrrrrrrKrFrrSrrfrlrbytesrrrrrrrrrarJr$r$r$r%s    $     ! @%$)CHN.   G  i      $   +     H!   #    7   /