o i^. @sddlZddlmZmZmZddlZddlmZddlmZm Z ddl m Z m ZGdddZdejfd d ZGd d d Zd ededededeedee f ddZdejfddZGdddZdS)N)IteratorListOptional)Tensor)_get_afilter_descStreamingMediaDecoder) CodecConfigStreamingMediaEncoderc@s.eZdZdZddZdefddZddZd S) _StreamingIOBufferz6Streaming Bytes IO buffer. Data are dropped when read.cCs g|_dSN)_buffer)selfrK/home/ubuntu/.local/lib/python3.10/site-packages/torchaudio/io/_effector.py__init__s z_StreamingIOBuffer.__init__bcCs|r|j|t|Sr )r appendlen)r rrrrwrites z_StreamingIOBuffer.writecCsV|jsdSt|jd|kr|jdS|jdd|}|jd|d|jd<|S)zMPop the oldest byte string. It does not necessary return the requested amountrN)r rpop)r nretrrrrs z_StreamingIOBuffer.popN)__name__ __module__ __qualname____doc__rbytesrrrrrrr s  r dtypec CJtjdtjdtjdtjdtjdi}||vr!td|d|||S)Nu8s16s32fltdblUnsupported dtype is provided . Supported dtypes are: torchuint8int16int32float32float64 ValueErrorkeysrtypesrrr_get_sample_fmt!sr2c@sFeZdZdZdededededeedeedefd d Z d d Z d S)_AudioStreamingEncoderz3Given a waveform, encode on-demand and return bytessrc sample_rateeffectmuxerencoder codec_configframes_per_chunkcCs\||_t|_t|j|d|_|jj|d|t|j|||d|j ||_ d|_ dSN)format) num_channelsr5r<r8 filter_descr9r) r4r buffer StreamWriterwriteradd_audio_streamsizer2ropenfpci_iter)r r4r5r6r7r8r9r:rrrr1s   z_AudioStreamingEncoder.__init__cCs|jjsA|jdkrA|jd|j|j|j|j|j|j7_|j|jdkr8|j|j d|_|jjsA|jdks |j |S)Nr) r@r rGrBwrite_audio_chunkr4rFrDflushcloser)r rrrrreadNs"   z_AudioStreamingEncoder.readN) rrrrrintstrrrrrLrrrrr3.s$ r3r4r5r6r7r8r9cCsxt}t||d}|j|d|t|j|||d||d|Wdn1s0wY| d|Sr;) ioBytesIOrArCrDr2rrErIseek)r4r5r6r7r8r9r@rBrrr_encodeYs   rRc Cr)Nr s16les32lef32lef64ler%r&r'r0rrr _get_muxerqsrWc @seZdZdZ  ddddddeedeedeedeed ef d d Zdd d Z dde de dee de fddZ dde de de dee de e f ddZdS) AudioEffectoraApply various filters and/or codecs to waveforms. .. versionadded:: 2.1 Args: effect (str or None, optional): Filter expressions or ``None`` to apply no filter. See https://ffmpeg.org/ffmpeg-filters.html#Audio-Filters for the details of filter syntax. format (str or None, optional): When provided, encode the audio into the corresponding format. Default: ``None``. encoder (str or None, optional): When provided, override the encoder used by the ``format``. Default: ``None``. codec_config (CodecConfig or None, optional): When provided, configure the encoding codec. Should be provided in conjunction with ``format`` option. pad_end (bool, optional): When enabled, and if the waveform becomes shorter after applying effects/codec, then pad the end with silence. Example - Basic usage To use ``AudioEffector``, first instantiate it with a set of ``effect`` and ``format``. >>> # instantiate the effector >>> effector = AudioEffector(effect=..., format=...) Then, use :py:meth:`~AudioEffector.apply` or :py:meth:`~AudioEffector.stream` method to apply them. >>> # Apply the effect to the whole waveform >>> applied = effector.apply(waveform, sample_rate) >>> # Apply the effect chunk-by-chunk >>> for chunk in effector.stream(waveform, sample_rate): >>> ... Example - Applying effects Please refer to https://ffmpeg.org/ffmpeg-filters.html#Filtergraph-description for the overview of filter description, and https://ffmpeg.org/ffmpeg-filters.html#toc-Audio-Filters for the list of available filters. Tempo - https://ffmpeg.org/ffmpeg-filters.html#atempo >>> AudioEffector(effect="atempo=1.5") Echo - https://ffmpeg.org/ffmpeg-filters.html#aecho >>> AudioEffector(effect="aecho=0.8:0.88:60:0.4") Flanger - https://ffmpeg.org/ffmpeg-filters.html#flanger >>> AudioEffector(effect="aflanger") Vibrato - https://ffmpeg.org/ffmpeg-filters.html#vibrato >>> AudioEffector(effect="vibrato") Tremolo - https://ffmpeg.org/ffmpeg-filters.html#tremolo >>> AudioEffector(effect="vibrato") You can also apply multiple effects at once. >>> AudioEffector(effect="") Example - Applying codec One can apply codec using ``format`` argument. ``format`` can be audio format or container format. If the container format supports multiple encoders, you can specify it with ``encoder`` argument. Wav format (no compression is applied but samples are converted to 16-bit signed integer) >>> AudioEffector(format="wav") Ogg format with default encoder >>> AudioEffector(format="ogg") Ogg format with vorbis >>> AudioEffector(format="ogg", encoder="vorbis") Ogg format with opus >>> AudioEffector(format="ogg", encoder="opus") Webm format with opus >>> AudioEffector(format="webm", encoder="opus") Example - Applying codec with configuration Reference: https://trac.ffmpeg.org/wiki/Encode/MP3 MP3 with default config >>> AudioEffector(format="mp3") MP3 with variable bitrate >>> AudioEffector(format="mp3", codec_config=CodecConfig(qscale=5)) MP3 with constant bitrate >>> AudioEffector(format="mp3", codec_config=CodecConfig(bit_rate=32_000)) NT)r8r9pad_endr6r<r8r9rYcCsB|dur|dus |durtd||_||_||_||_||_dS)NzM`encoder` and/or `condec_config` opions are provided without `format` option.)r.r6r<r8r9rY)r r6r<r8r9rYrrrrs  zAudioEffector.__init__cCs|j\}}|jdur|j}|j}i} |jdkr||d} nt|j}d}||d} |dur=t|||j|||j} n t|||j|||j|} |durO|n|} t | t |j|} |j rd| d|} t | || d} | j |ppdd| d| S)Nmulaw)r5channelsz,apad=whole_len=)r<optionrH)r?)shaper<r8rWrrRr6r9r3rr2rY StreamReaderrC)r waveformr5output_sample_rater: num_framesr>r7r8r\r4 output_srr?readerrrr _get_readers.    zAudioEffector._get_readerr_r5r`returncCsR|jdkr td|j|dkr|S||||}||\}t|S)aApply the effect and/or codecs to the whole tensor. Args: waveform (Tensor): The input waveform. Shape: ``(time, channel)`` sample_rate (int): Sample rate of the input waveform. output_sample_rate (int or None, optional): Output sample rate. If provided, override the output sample rate. Otherwise, the resulting tensor is resampled to have the same sample rate as the input. Default: ``None``. Returns: Tensor: Resulting Tensor. Shape: ``(time, channel)``. The number of frames could be different from that of the input. -Expected the input waveform to be 2D. Found: r)ndimr.numelrdprocess_all_packets pop_chunksr)r r_r5r`rcappliedrrrapply"s   zAudioEffector.applyr:ccsZ|jdkrtd|j|dkr|S|||||}|D]\}t|Vq"dS)aZApply the effect and/or codecs to the given tensor chunk by chunk. Args: waveform (Tensor): The input waveform. Shape: ``(time, channel)`` sample_rate (int): Sample rate of the waveform. frames_per_chunk (int): The number of frames to return at a time. output_sample_rate (int or None, optional): Output sample rate. If provided, override the output sample rate. Otherwise, the resulting tensor is resampled to have the same sample rate as the input. Default: ``None``. Returns: Iterator[Tensor]: Series of processed chunks. Shape: ``(time, channel)``, where the the number of frames matches ``frames_per_chunk`` except the last chunk, which could be shorter. rfrgrN)rhr.rirdstreamr)r r_r5r:r`rcrlrrrrn>s   zAudioEffector.stream)NNr )rrrrrrNrboolrrdrrMrmrrnrrrrrXsBr  rX)rOtypingrrrr(r!torio.io._streaming_media_decoderrrr^!torio.io._streaming_media_encoderrr rAr rr2r3rMrNrRrWrXrrrrs0  +