o pi9 @sdZddlZddlZddlmZddlmZddlmZm Z m Z ddl m m ZddlZddlmZddlmZzddlZddlmZdd lmZmZWneygZzed eWYdZ[ndZ[wweeBeBeBZd Zd ed dfddZGdddZ dS)z] # Audio IO pyannote.audio relies on torchcodec for reading and torchaudio for resampling. N)IOBase)Path)MappingOptionalTuple)Segment)Tensor) AudioSamples) AudioDecoderAudioStreamMetadataz torchcodec is not installed correctly so built-in audio decoding will fail. Solutions are: * use audio preloaded in-memory as a {'waveform': (channel, time) torch.Tensor, 'sample_rate': int} dictionary; * fix torchcodec installation. Error message was: aI Audio files can be provided to the Audio class using different types: - a "str" or "Path" instance: "audio.wav" or Path("audio.wav") - a "IOBase" instance with "read" and "seek" support: open("audio.wav", "rb") - a "Mapping" with any of the above as "audio" key: {"audio": ...} - a "Mapping" with both "waveform" and "sample_rate" key: {"waveform": (channel, time) torch.Tensor, "sample_rate": 44100} For last two options, an additional "channel" key can be provided as a zero-indexed integer to load a specific channel: {"audio": "stereo.wav", "channel": 0} filereturnr cCs.t|dj}t|dtr|dd|S)a6Protocol preprocessor used to cache audio metadata This is useful to speed future random access to this file, e.g. in dataloaders using Audio.crop a lot.... Parameters ---------- file : AudioFile Returns ------- metadata : AudioStreamMetadata Audio file metadata audior)r metadata isinstancerseek)r rrJ/home/ubuntu/.local/lib/python3.10/site-packages/pyannote/audio/core/io.pyget_audio_metadataFsrc seZdZdZdZededefddZedede fdd Z dd e ffd d Z dded e de d Bde ee ffddZdedefddZ dded ee de fddZdede ee ffddZ ddedede ee ffddZZS)AudioaAAudio IO Parameters ---------- sample_rate: int, optional Target sampling rate. Defaults to using native sampling rate. mono : {'random', 'downmix'}, optional In case of multi-channel audio, convert to single-channel audio using one of the following strategies: select one channel at 'random' or 'downmix' by averaging all channels. Usage ----- >>> audio = Audio(sample_rate=16000, mono='downmix') >>> waveform, sample_rate = audio({"audio": "/path/to/audio.wav"}) >>> assert sample_rate == 16000 >>> sample_rate = 44100 >>> two_seconds_stereo = torch.rand(2, 2 * sample_rate) >>> waveform, sample_rate = audio({"waveform": two_seconds_stereo, "sample_rate": sample_rate}) >>> assert sample_rate == 16000 >>> assert waveform.shape[0] == 1 gMbP?waveformr cCs"|jddd}||dS)zPower-normalize waveform Parameters ---------- waveform : (..., time) Tensor Waveform(s) Returns ------- waveform: (..., time) Tensor Power-normalized waveform(s) Tdimkeepdimg:0yE>)squaremeansqrt)rrmsrrrpower_normalizeys zAudio.power_normalizer cCs t|trn t|ttfrt|t|jd}nt|tr"|ddSttd|vrY|d}t|j dks?|j d|j dkrCtd| dd }|d urQtd | d d|Sd |vrt|d trf|St|d }| sxtd |d| d |j|Std)aValidate file for use with the other Audio methods Parameter --------- file: AudioFile Returns ------- validated_file : Mapping {"audio": str, "uri": str, ...} {"waveform": tensor, "sample_rate": int, "uri": str, ...} {"audio": file, "uri": "stream"} if `file` is an IOBase instance Raises ------ ValueError if file format is not valid or file does not exist. )ruristreamrrz>'waveform' must be provided as a (channel, time) torch Tensor. sample_rateNz5'waveform' must be provided with their 'sample_rate'.r rzFile z does not existz:Neither 'waveform' nor 'audio' is available for this file.) rrstrrstemr ValueErrorAudioFileDocStringlenshapeget setdefaultis_file)r rr$pathrrr validate_files>   "   zAudio.validate_fileNr$cst||_||_dSN)super__init__r$mono)selfr$r3 __class__rrr2s  zAudio.__init__channelcCs|dur |||d}|jd}|dkr7|jdkr+td|d}|||d}n |jdkr7|jddd}|jdurM|j|krMtj|||j}|j}||fS)aDownmix and resample Parameters ---------- waveform : (channel, time) Tensor Waveform. sample_rate : int Sample rate. channel : int, optional Channel to use. Returns ------- waveform : (channel, time) Tensor Remixed and resampled waveform sample_rate : int New sample rate Nr#rrandomdownmixTr) r*r3r8randintrr$ torchaudio functionalresample)r4rr$r7 num_channelsrrrdownmix_and_resamples   zAudio.downmix_and_resamplecCs>||}d|vrt|dj}|d}||St|}|jS)zGet audio file duration in seconds Parameters ---------- file : AudioFile Audio file. Returns ------- duration : float Duration in seconds. rr$)r/r)Trduration_seconds_from_header)r4r framesr$rrrr get_durations zAudio.get_durationdurationcCs&|p|j}|dur tdt||S)z=Deterministic number of samples from duration and sample rateNz<`sample_rate` must be provided to compute number of samples.)r$r'round)r4rDr$rrrget_num_sampless  zAudio.get_num_samplescCs||}|dd}d|vr|d}|d}|j|||dSt|d}|}|j}|j}t|dtr=|d d|j|||dS)a%Obtain waveform Parameters ---------- file : AudioFile Returns ------- waveform : (channel, time) torch.Tensor Waveform sample_rate : int Sample rate See also -------- AudioFile r7Nrr$r7rr) r/r+r?r get_all_samplesdatar$rrr)r4r r7rr$decodersamplesrrr__call__#s   zAudio.__call__raisesegmentc Cs||}|dd}d|vr|d}|j\}}|d}||} ||j|} td| } | dkrA|dkr?td|jdd d} ||j|} t| ||} | |krn|dkrltd |jdd |d d d| dd|} |dd| | f}t || | f}|j |||dSt |d}|j }|j }|j} ||j|}t|j}t|j}td|| |} |dkr|dkrtd|dd d}t|||||} || kr|dkrtd |dd |d d d| dd| }|||}|j}|j }t|dtr |dd||j|}|j\}}| || |}t|dkr9td|d|d d d|d|d |dkrI|ddddf}n |dkrR| d7} t || | f}|j |||dS)a6Fast version of self(file).crop(segment, **kwargs) Parameters ---------- file : AudioFile Audio file. segment : `pyannote.core.Segment` Temporal segment to load. mode : {'raise', 'pad'}, optional Specifies how out-of-bounds segments will behave. * 'raise' -- raise an error (default) * 'pad' -- zero pad Returns ------- waveform : (channel, time) torch.Tensor Waveform sample_rate : int Sample rate r7Nrr$rrMz,requested chunk with negative start time (t=z.3fzs)z!requested chunk with end time (t=zs) greater than r z in-memoryz file duration (zs).rGrgr#zrequested chunk z from z file resulted in z! samples instead of the expected z samples.r)r/r+r*rFstartmaxr'endFpadr?r rr$rAfloatget_samples_played_in_rangerIrrrrDabs)r4r rNmoder7r_ num_samplesr$rD start_sample pad_start end_samplepad_endrIrJrrOrQrKexpected_num_samplesactual_num_samples differencerrrcropKs                z Audio.crop)NNr0)rM)__name__ __module__ __qualname____doc__ PRECISION staticmethodrr AudioFilerr/intr2rr?rTrCrrFrLrra __classcell__rrr5rr_sH@  , , r)!rer8warningsiorpathlibrtypingrrrtorch.nn.functionalnnr<rRr; pyannote.corertorchr torchcodecr torchcodec.decodersr r Exceptionewarnr%rhr(rrrrrrs4