o }oi`@sddlZddlZddlZddlZddlmZmZmZmZm Z m Z ddl Z ddl Z ddl mZddlZddlZddlmZddlmZmZddlmZddlmZddlmZddlmZdd l m!Z!m"Z"dd l#m$Z$m%Z%m&Z&m'Z'm(Z(dd l)m*Z*zddl+Z,d Z-Wn e.yd Z-Ynwde/de e0ee0fde1fddZ2de0de0fddZ3Gddde4Z5 drde6dee0de0deee0fddZ7Gddde4Z8de6de6fd d!Z9d"e6d#e5d$eee0d%e/de6f d&d'Z:d(e/d)ee/eej;fd*e6fd+d,Ze0ffd0d1Z?d2e e6e0ej>fde e6e0ej>ffd3d4Z@d-eeej>deej>fd5d6ZAdtd(e/d7e/fd8d9ZBGd:d;d;e4ZCdd-ej>dej>fd=d>ZDdud-ej>d@e0dAee=dee0fdBdCZEdvdEej>dFe0dGe0dej>fdHdIZF J K LdwdEej>dMej>d@e0dNe0dOe0dPe0de e0e0ffdQdRZG  DdxddSej>dTe0d@e0dUe=dGe0dej>fdVdWZH     XdydYe/d@e=dZee6d[ee/d\ee=d]eej>d^eej>d_e=deIfd`daZJd(e/d@e0fdbdcZK d Ddzd@e=dee6dfe6dge6dhe6die/dje0dGe0de6fdkdlZLdmeIde6fdndoZMdtd(e/d7e/fdpdqZNdS){N)DictIterableListOptionalTupleUnion) default_rng) DictConfig OmegaConf)convolve)Rotation)tqdm) AudioSegment) read_manifestwrite_manifest)db2mag generate_approximate_noise_fieldmag2dbpow2dbrms)loggingTFkeyvalreturncCs t|r |}}nt|}t|}|dkr'd|kr&|kr&dkr'dS|dkrr(r(r(r)r|s    OMG rkwargscCstdi|S)aZWrapper around `simulate_room` to handle kwargs. `pool.map(simulate_room_kwargs, examples)` would be equivalent to `pool.starstarmap(simulate_room, examples)` if `starstarmap` would exist. Args: kwargs: kwargs that are forwarded to `simulate_room` Returns: Dictionary with metadata, see `simulate_room` Nr()r.)r@r(r(r)r-s r-rrrr cCstj|d|dt|d|dd}tj|d|dt|d|dd}||fD]}||jj|D]}||q7|q,g}g} g} |D]}||\} } } | | | | | | qLt |j t |j d}id |d|dd|dd |dd |dd | d | jddd| d| jddd|dd|dd|jd|jd|d|d| d| dt|i}t|||t|S)aSimulate room Args: room_params: parameters of the room to be simulated mic_array: defines positions of the microphones source_positions: positions for all sources to be simulated room_filepath: results are saved to this path Returns: Dictionary with metadata based on simulation setup and simulation results. Used to create the corresponding manifest file. rrrr)fs materialsrrr)rirrr rir_absorption rir_max_orderrir_rt60_theoryrir_rt60_measuredrr6anechoic_rt60_theoryanechoic_rt60_measuredr0 mic_centerrsource_distancesource_azimuthsource_elevation num_sources)rShoeBoxMaterialadd_microphone_arrayrErG add_source compute_rirrarconvert_rir_to_multichannelrC rt60_theory measure_rt60r<r1rsave_rir_simulationconvert_numpy_to_serializable)rrrr room_sim room_anechoicrrrKrLrMr`rr rir_datasetr:r(r(r)r.s              r.filepathr[r:c Cstj|r td||d}t|dU}|D]2\}}t||kr6td|d|dt|d| |}t |D]\}} |j |||dq?q| d } |D] \} } | j | | dqXWd d S1sowYd S) zSave simulated RIRs and metadata. Args: filepath: Path to the file where the data will be saved. rir_dataset: Dictionary with RIR data. Each item is a set of multi-channel RIRs. metadata: Dictionary with related metadata. zOutput file exists: rNwz%Each RIR dataset should have exactly z elements. Current RIR z has z elements)r<r:N) r!r"r$rh5pyFiler rr% create_grouprcreate_dataset) r\r[r:rNh5frir_key rir_value rir_grouprrCmetadata_grouprvaluer(r(r)rWis$    "rWrCrrccCs`t|d}|||dd}|ddd}Wd||fS1s'wY||fS)aHLoad simulated RIRs and metadata. Args: filepath: Path to simulated RIR data source: Index of a source. rir_key: String to denote which RIR to load, if there are multiple available. Returns: Multichannel RIR as ndarray with shape (num_samples, num_channels) and scalar sample rate. rNr:rr()r^r_)r\rrcrbrCrr(r(r)load_rir_simulations  rir<cCst|tr|D] \}}t|||<q |St|tr$dd|D}|St|tjr0|}|St|tjrz1convert_numpy_to_serializable..)r8rr rXr+r!rtolistintegerintfloatingrgenericitem)r<rrr(r(r)rXs&      rXcst}td}dg|}t|D]2fddt|D}t|}t||f|<t|D]\}}||d||f<q2q|S)zConvert RIR to a list of arrays. Args: rir: list of lists, each element is a single-channel RIR Returns: List of multichannel RIRs rNcsg|] }t|qSr()r)rjrn_sourcerCr(r)rlsz/convert_rir_to_multichannel..)rrr$r!rr)rCrCrNmc_rirrir_lenmax_lenn_miclen_micr(rtr)rTs   rTrc Cst|}g}g}g}g}g}g}g} g} |D]9} || d7}|| d7}|| d7}|dd| dD7}|| d|| d7}| | d | | d 7} qtjd d td ddtj|ddtdtdtdtd dd tj|ddtdtdtdtd ddtj|ddtdtdtdtd ddtj|ddtdtdtdtd ddtj|d dtd!tdtd"td dd#tj|d$dtd!tdtd%td dd&tj| d dtd!tdtd'td dd(tj| d$dtd!tdtd)t d(D]} td d| dt tj d*d+qBt |d,urst |ttd-|d,Sd,S).zPlot distribution of parameters from manifest file. Args: filepath: path to a RIR corpus manifest file plot_filepath: path to save the plot at rKrLrMcSsg|]}|dqS)r3r()rjrr(r(r)rlrmz*plot_rir_manifest_info..rrFrGrHrI figsizer3r4r`ry distance / m # exampleszSource-to-array center distancer azimuth / degzSource-to-array center azimuthr5relevation / degz Source-to-array center elevationz source heightz height / mz Source heighttheoryRT60 / sz RT60 theoryr}measuredz RT60 measuredzRT60 theory (anechoic)zRT60 measured (anechoic) lower leftlocNPlot saved at %s)rrrrsubplothistxlabelylabeltitlergridr tight_layoutsavefigcloserr) r\rr:rKrLrM source_heightrFrGrHrIr<rr(r(r)r1s                                r1c@seZdZdZdefddZeddZeddZej d dZd d Z d e d e fddZ d e de d ee fddZd e de d e fddZddZdS)RIRMixGeneratoraCreates a dataset of mixed signals at the microphone by combining target speech, background noise and interference. Correspnding signals are are generated and saved using the `generate` method. Input configuration is expexted to have the following structure ``` sample_rate: sample rate used for simulation room: subset: manifest for RIR data target: subset: manifest for target source data noise: subset: manifest for noise data interference: subset: manifest for interference data interference_probability: probability that interference is present max_num_interferers: max number of interferers, randomly selected between 0 and max mix: subset: num: number of examples to generate rsnr: range of RSNR rsir: range of RSIR ref_mic: reference microphone ref_mic_rms: desired RMS at ref_mic ``` rc Cstd||_||jj|_tdt|jt |jt |_ |jD]S}t }td|dD]@}zt |j||||<tdd|t||Wq4t yt}zd||<tdd|td t |WYd}~q4d}~ww||j |<q'td |jd d |_dS) z Instantiate a RIRMixGenerator object. Args: cfg: generator configuration defining data for room, target signal, noise, interference and mixture zInitialize RIRMixGeneratorzInitialized with %d subsets: %szLoading data for %s)rtargetnoise interferencez %-*s: %d filesNz %-*s: 0 filesz) Manifest data not loaded. Exception: %szLoaded all manifestsrr)rrrrrrkeyssubsetsrrrr:rrrrr)r?rr3 subset_datarrr(r(r)rATs,     zRIRMixGenerator.__init__cCrrrrBr(r(r)rxrzRIRMixGenerator.cfgcCrrrrBr(r(r)rrzRIRMixGenerator.sample_ratecCrRrrrr(r(r)rrTc Cs|jd}|durtd|dkrtd||jd}|s%td|jd}|s1td d D])}||}|dusCt|rDq3t|d ksR|d|d ks\td |d|q3|jd}|sitd|jd}|swtdn@|dd}|dd} |dd} |dur|dkrtd||dkr| dur| dksJd| | dur| dksJd|jd} | stdd| vrtdd| vrtddS)rrNrrzSample rate must be positive: rz]Room configuration not provided. Expecting RIR manifests in format {subset: path_to_manifest}rzaTarget configuration not provided. Expecting audio manifests in format {subset: path_to_manifest}rrr`r3r4rrrz`Noise configuration not provided. Expecting audio manifests in format {subset: path_to_manifest}rz(Interference configuration not provided.interference_probabilitymax_num_interferersmin_azimuth_to_targetz>Interference probability must be non-negative. Current value: z;Max number of interferers must be positive. Current value: z*Min azimuth to target must be non-negativemixzHMix configuration not provided. Expecting configuration for each subset.ref_micz!Reference microphone not defined. ref_mic_rmsz%Reference microphone RMS not defined.)rrr%r!r"rrr) r?rr target_cfgrrg noise_cfginterference_cfgrrrmix_cfgr(r(r)rsj           zRIRMixGenerator.check_cfgr3rc s,fdd}j|d}jjt|jdd}|||\}}|dur-tdjj||}t|d}d jvrjjd } jjd } | d kr\t d | jj|| dd} n>t | t rt dt | dt| kru|ksnJd|d| t| } ntd| t d||} t| } t| | ksJd| dt| d| t dt | t|d| } t|d|} tjj| | dd}|d}tj|stjjj|}tj||}t||||d|| | | |d||d||d||d }|S)!a Prepare a dictionary with target configuration. The output dictionary contains the following information ``` room_index: index of the selected room from the RIR corpus room_filepath: path to the room simulation file source: index of the selected source for the target rt60: reverberation time of the selected room num_mics: number of microphones azimuth: azimuth of the target source, relative to the microphone array elevation: elevation of the target source, relative to the microphone array distance: distance of the target source, relative to the microphone array audio_filepath: path to the audio file for the target source text: text for the target source audio signal, if available duration: duration of the target source audio signal ``` Args: subset: string denoting a subset which will be used to selected target audio and room parameters. Returns: Dictionary with target configuration, including room, source index, and audio information. c s|D]a}||}jj|djdd}|D]L}g}dD]9}jj|durU|d||}jj|d|krFjj|dkrNnn|d q|dnqt|rb||fSqqd S) z8Find a room and a source that satisfies the constraints.rNFsizereplacerNsource_rr4Tr)rchoicerrrrrr) room_metadata room_indices room_index room_datasourcesrconstraints_met constraint source_valuerBr(r)select_target_sources$0   z=RIRMixGenerator.generate_target..select_target_sourcerFrNz:Could not find a feasible source given target constraints r0rrC selectionrzRandomly selecting %d micsz"Using explicitly selected mics: %srz$Expecting mic_selection in range [0,z), current value: z$Unexpected value for mic_selection: zUsing all %d available micsz Expecting z mics, but received z mics: zSelected mics: %srr4r6r rGrLrMrK) rr rr selected_micsrr0rrr`distance_source_to_mic)r:rrrrrrrrrrr8rrr#r!r9r%arangerIrJr!r"isabsdirnamerr#rprn)r?r3rrrrrrnum_available_micsrC mic_selectionrr0rrr  manifest_dirrr(rBr)generate_targets`                zRIRMixGenerator.generate_targetrcCs|j|ddur dS|jjdd}|jjdd}|dkr8|dkr8|jjddd |kr8|jjd|dd }ndS|d }|j|d |}tt|d }| |d |jjdd} g} t|D]s} d} t |dkr| dur|j |} | | | dkr|d| } t | |d}t || krd} qht |dkr| dusr| durtd| || r| SdS| |d|d| |d| |d| |d| d}| |qd| S)a Prepare a list of dictionaries with interference configuration. Args: subset: string denoting a subset which will be used to select interference audio. target_cfg: dictionary with target configuration. This is used to determine the minimal required duration for the noise signal. Returns: List of dictionary with interference configuration, including source index and audio information for one or more interference sources. rNrrrr4grwrrrrNrrrLrz8Could not select a feasible interference source %d of %srrrMrK)rrpositionrrr`)r:rrrrrintegersr+rremoverrr.absrrr)r?r3rmax_num_sourcesrnum_interferersrrfeasible_sourcesrrrrrL azimuth_diffinterfering_sourcer(r(r)generate_interferenceXsN           z%RIRMixGenerator.generate_interferencecCst}dD]M}||jj|vr|jj||}n|jj|}|dur)d||<qt|r3|||<qt|dkrJ|jj|d|ddd||<q|j |||<q|ddkrbt |d |d<d |jjvrqt |jjd ni|d <|S) aGenerate scaling parameters for mixing the target speech at the microphone, background noise and interference signal at the microphone. The output dictionary contains the following information ``` rsnr: reverberant signal-to-noise ratio rsir: reverberant signal-to-interference ratio ref_mic: reference microphone for calculating the metrics ref_mic_rms: RMS of the signal at the reference microphone ``` Args: subset: string denoting the subset of configuration target_cfg: dictionary with target configuration Returns: Dictionary containing configured RSNR, RSIR, ref_mic and RMS on ref_mic. )rsnrrsirrr min_durationNr3rr4rrclosestrr2) rrrrr!r"rrrrargminr to_object)r?r3rrrrgr(r(r) generate_mixs     " &zRIRMixGenerator.generate_mixc Cstdt|jjd|_|jj}|dr|dd}td||jD]}t j ||}g}t j |sCtd|t |nt j |rZtt |dkrZtd |d |jj|j}td ||tt||d |d D].}||}|||}|||} t j ||d|d} |j||| | d} || qt|j|dt j |jj||j|dt j |jj|d} |dur| |j|dt j |jj |d|j!d} durt "d} | dur5| dkr5td| dt#|t$%| t|}t&j'| d}t(t|)t*|t|d|d }Wdn 1s/wYn%tdg}t|t|d|d D]} |t+d+i| d| iqHt j |t j ,|d|d}t j |rt j -|rtd |t|t|d!|d D]"}|.D]\}}|d"r|durt j j/||d#||<qqt0||t j |t j ,|d|d$}t j |rt j -|rtd%|t1||d&q%t j |d'}t j |rt j -|rtd(|t2j3|j|d)d*dS),aGenerate a corpus of microphone signals by mixing target, background noise and interference signals. This method will prepare randomized examples based on the current configuration, run simulations and save results to output_dir. zGenerate mixed signalsrrNrrrrrrz#Preparing %d examples for subset %sz Preparing )rrf _example_09d)rrrrbase_output_filepathrr)r target_dirr noise_dirr)rinterference_dirr r4r r rz Simulating raudio_metadata_z.jsonrzMaking filepaths relative _filepathrrrrrrTrr()4rrrrrrrrrr!r"r#r$r%r&rr'rrrr rrrrrrr:rrrupdaterrr(zip itertoolsrepeatr)r*r+r,simulate_room_mix_helpersimulate_room_mixbasenamer/r r0rplot_mix_manifest_infor r2)r?rr3r5r6 num_examples n_examplerrrrr8rr examples_and_audio_metadatar9r:r;r<rrrr=r(r(r)r>s                   " "zRIRMixGenerator.generateN)rrrrr rArrrr?rrrrrrrr>r(r(r(r)r6s$    IT 3rsignalcCst|}|jdkrt||d|}|S|jdkrC|jd}t||f}t|D]}t||dd|fd||dd|f<q)|Std|jd)aConvolve signal with a possibly multichannel IR in rir, i.e., calculate the following for each channel m: signal_m = rir_m st signal Args: signal: single-channel signal (samples,) rir: single- or multi-channel IR, (samples,) or (samples, channels) Returns: out: same length as signal, same number of channels as rir, shape (samples, channels) r4Nr3z RIR with z not supported)rr:r r;r!rrr)rrC num_samplesout num_channelsrr(r(r) convolve_rirZs   ,r@rn_directc Cst||d}|j\}}dg|}t|D]M}t|||d} |||} tt|| | |fd|} tt|d| |fdtt|| d|fd|} t| | ||<q|S)a5Calculate direct-to-reverberant ratio (DRR) from the measured RIR. Calculation is done as in eq. (3) from [1]. Args: rir: room impulse response, shape (num_samples, num_channels) sample_rate: sample rate for the impulse response n_direct: direct path delay n_0_ms: window around n_direct for calculating the direct path energy Returns: Calculated DRR for each channel of the input RIR. References: [1] Eaton et al, The ACE challenge: Corpus description and performance evaluation, WASPAA 2015 Nrr3)rpr;rr$r!sumrr) rCrrn_0_msn_0len_rirrdrrr dir_startdir_endpow_dirpow_reverberantr(r(r) calculate_drrws    $Dr缉ؗҜt|t|krtdt|dt|t||d}t|}t|}t|} d} } } tdt|||D]=} t| | |}t||}t| |}||krw||krw| t t ||d7} | t t ||d7} | |7} q:| t||krtd|d| |d| | } | | } t | t | fS) a+Calculate RMS over segments where both input signals are active. Args: x: first input signal y: second input signal sample_rate: sample rate for input signals in Hz rms_threshold_db: threshold for determining activity of the signal, relative to max absolute value window_len_ms: window length in milliseconds, used for calculating segmental RMS min_active_duration: minimal duration of the active segments Returns: RMS value over active segments for x and y. z)Expecting signals of same length: len(x)=z , len(y)=rrr3z,Signals are simultaneously active less than z s: only z s) rrrprrrslicerr!rrsqrt)rrrrrr window_len rms_threshold x_normalized y_normalizedx_active_powery_active_power active_lenrwindow x_window_rms y_window_rmsr(r(r)simultaneously_active_rmss.   r disturbancesdr ref_channelc Csp|j|jkrtd|jd|jt|dd|f|dd|f|d\}}t| |||}||} | S)a Args: signal: numpy array, shape (num_samples, num_channels) disturbance: numpy array, same shape as signal sdr: desired signal-to-disturbance ration sample_rate: sample rate of the input signals ref_channel: ref mic used to calculate RMS eps: regularization constant Returns: Scaled disturbance, so that signal-to-disturbance ratio at ref_channel is approximately equal to input SDR during simultaneously active segment of signal and disturbance. z,Signal and disturbance shapes do not match: z != Nr)r;r%rr) rrrrrr signal_rmsdisturbance_rmsdisturbance_gainscaled_disturbancer(r(r)rs  rrt signal_type audio_data audio_dirr ref_signalr0rc Cs|dvr td|d|durdSi}|durd} |dur$t||nd} ggggd} | dkrt|} | d} tj| sK|durKtj|| } t| |t j | || d | d dd }|d krt|j dkrst d |j d| ntd|d| d| | d | d | d | d d| d| dt|j}| durt| |fn|} | t|8} | dks1t| d dd}tdd| dDrd| d|d<n\t|}t|D]R}d} gggd} |d kr|} n |dkrt|}||} | dkrt|} | d} tj| s$|dur$tj|| } t| || d t| |}dkrBtd|}d}nd| d }}t j | |||d }|j dkr\|j}ntt|j }|jdd|f}| d| | d t||| d |t|}| durt| |fn|} | t|8} | dks|dkr|dkrz| d|||dj} tt|| |d} Wn%ty}ztdt |tddi} } WYd}~qd}~ww| d|df} |j!dkr| j!dkr| dddf} z t"|| |dWn%ty7}ztdt |tddi} } WYd}~qd}~ww| durCt#d|| |d <| |fS)!a5Prepare an audio signal for a source. Args: signal_type: 'point' or 'diffuse' sample_rate: Sampling rate for the signal audio_data: List of audio items, each is a dictionary with audio_filepath, duration, offset and optionally text audio_dir: Base directory for resolving paths, e.g., manifest basedir min_duration: Minimal duration to be loaded if ref_signal is not provided, in seconds ref_signal: Optional, used to determine the length of the signal mic_positions: Optional, used to prepare approximately diffuse signal num_retries: Number of retries when selecting the source files Returns: (audio_signal, metadata), where audio_signal is an ndarray and metadata is a dictionary with audio filepaths, durations and offsets )rZdiffusezUnexpected signal type rNr4)audio_filepathdurationoffsetrrrr r!) audio_file target_srr r!rZz5Expecting single-channel source signal, but received z. File: r)r r!cSsg|]}t|tqSr()r8r)rjttr(r(r)rl`sz)prepare_source_signal.. )rr r!r)r0 noise_signalrz.Failed to generate approximate noise field: %sz Try again..r3rz Exception: %sz'Signals are not overlapping, try again.zAudio signal not set: %s.source_signals)$r%rprrr!r"rr#check_min_sample_rater from_filerrrrrsamplesr! concatenaterrrrceilrreshaperGrr9rrrrr:rr)rrrrrrr0rr: audio_signalsamples_to_loadsource_signals_metadatarsr audio_segmentsegment_samples total_lenrrC max_offsetr!r selected_channelrr(r(r)prepare_source_signal s       )          ,          r7cCs4tj|d}||krtd|d|d|ddS)aMake sure the file's sample rate is at least sample_rate. This will make sure that we have only downsampling if loading this file, while upsampling is not permitted. Args: filepath: path to a file sample_rate: desired sample rate )r"z Sample rate (z)) is lower than the desired sample rate ( ). File: rN)librosaget_samplerater)r\rfile_sample_rater(r(r)r)s r)+?rrrrr max_amplitudec'Cs d?dtdtdtdtdtdtjf dd } d@dtjd tjdtd tdtjf d d}    dAdtdtdttjdtdtdttdtdtfdd} ||d|d|d|d} ||d|d||ddd} | | | |d} td ||d!|d"|d#d$\}}d!|d%i}t|| }t|| }t|| }td&||d'|d(|d)|d*\}}|d%|d(<|durd}n:d+}g|d,<|D]1}td ||d,|d-|d.\}}|d, |d%||d|d|d|d}t||}||7}q| }|durt |||d/||dd0}||7}|durt |||d1||dd0}||7}t |dd|df}t |d2||}tt|}||} |kr\|| }!||!9}|d2t|!7<td3|t|!||||||d4}"i}#|"D]?\}$}%|%durx||%}%| ||$|%||d|$d|d|ddd|dddd5}&|$d6kr|&|#d7<qk|&|#|$d8<qk|#|d9|d:||||d|d;t| |tj| d+d<d=|durdn|d/|durdn|d1|d> t|#S)BaSimulate mixture signal at the microphone, including target, noise and interference signals and mixed at specific RSNR and RSIR. Args: sample_rate: Sample rate for all signals target_cfg: Dictionary with configuration of the target. Includes room_filepath, source index, audio_filepath, duration noise_cfg: List of dictionaries, where each item includes audio_filepath, offset and duration. interference_cfg: List of dictionaries, where each item contains source index mix_cfg: Dictionary with the mixture configuration. Includes RSNR, RSIR, ref_mic and ref_mic_rms. audio_metadata: Dictionary with a list of files for target, noise and interference base_output_filepath: All output audio files will be saved with this prefix by adding a diffierent suffix for each component, e.g., _mic.wav. max_amplitude: Maximum amplitude of the mic signal, used to prevent clipping. eps: Small regularization constant. Returns: Dictionary with metadata based on the mixture setup and simulation results. This corresponds to a line of the output manifest file. rCr rrrrcrcSsDt|||d\}}||krtd|d|d||dd|fS)aLoad a RIR and check that the sample rate is matching the desired sample rate Args: room_filepath: Path to a room simulation in an h5 file source: Index of the desired source sample_rate: Sample rate of the simulation rir_key: Key of the RIR to load from the simulation. Returns: Numpy array with shape (num_samples, num_channels) )rrczRIR sample rate (z,) is not matching the expected sample rate (r8N)rir)r rrrrcrCrir_sample_rater(r(r)load_rirs z#simulate_room_mix..load_rir皙? rir_anechoicearly_durationcSsDt||}ttj|dd}|}d|||dddf<|S)z&Return only the early part of the RIR.rr6N)rpr!r#argmaxcopy)rCrArrB early_lendirect_path_delay rir_earlyr(r(r) get_early_rirs z(simulate_room_mix..get_early_rirrNwavr base_pathtagr/r2rformatsubtypec SsP|dus|sdS|dkr|dd|f}|d|d|}t|||||S)z&Save audio signal and return filepath.Nrrr)sfwrite) rJrKr/rr2rrLrMrr(r(r) save_audios z%simulate_room_mix..save_audio)rrrr)rrrrc)rCrArrZrrr)rrrrrr(rr0rr)rrr0rrrrrr)rrrrrr)rrrrrrrzs6         "+ f  4  $[&!*"c(& (  < ' C