o 8wiQ4@sddlZddlmZmZddlmZddlmZmZm Z m Z m Z ddl Z ddlZddlmZddlmZddlmZdd lmZmZdd lmZmZerXdd lmZd d lmZeGdddZ ddde!fddZ"dS)N) dataclassfield)BytesIO) TYPE_CHECKINGAnyClassVarOptionalUnion)config)DownloadConfig) array_cast) is_local_pathxopen)no_op_if_value_is_nullstring_to_dict AudioDecoder) FeatureTypec @sNeZdZUdZdZeeed<dZe ed<dZ eeed<e dddZ ee ed <d Zee ed <eeed Zeeed <e ddddZe ed<ddZdee eeedfdefddZ d"dedeee ee e dffddfddZdedee dfffddZdeejej fdej fddZ!d"dej dej fd d!Z"dS)#AudioaJAudio [`Feature`] to extract audio data from an audio file. Input: The Audio feature accepts as input: - A `str`: Absolute path to the audio file (i.e. random access is allowed). - A `dict` with the keys: - `path`: String with relative path of the audio file to the archive file. - `bytes`: Bytes content of the audio file. This is useful for parquet or webdataset files which embed audio files. - A `dict` with the keys: - `array`: Array containing the audio sample - `sampling_rate`: Integer corresponding to the sampling rate of the audio sample. - A `torchcodec.decoders.AudioDecoder`: torchcodec audio decoder object. Output: The Audio features output data as `torchcodec.decoders.AudioDecoder` objects, with additional keys: - `array`: Array containing the audio sample - `sampling_rate`: Integer corresponding to the sampling rate of the audio sample. Args: sampling_rate (`int`, *optional*): Target sampling rate. If `None`, the native sampling rate is used. mono (`bool`, defaults to `True`): Whether to convert the audio signal to mono by averaging samples across channels. decode (`bool`, defaults to `True`): Whether to decode the audio data. If `False`, returns the underlying dictionary in the format `{"path": audio_path, "bytes": audio_bytes}`. stream_index (`int`, *optional*): The streaming index to use from the file. If `None` defaults to the "best" index. Example: ```py >>> from datasets import load_dataset, Audio >>> ds = load_dataset("PolyAI/minds14", name="en-US", split="train") >>> ds = ds.cast_column("audio", Audio(sampling_rate=44100)) >>> ds[0]["audio"] >>> audio = ds[0]["audio"] >>> audio.get_samples_played_in_range(0, 10) AudioSamples: data (shape): torch.Size([2, 110592]) pts_seconds: 0.0 duration_seconds: 2.507755102040816 sample_rate: 44100 ``` N sampling_rateTdecode stream_indexF)defaultrepriddictdtypebytespathpa_type)rinitr_typecCs|jSN)r")selfr'T/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/datasets/features/audio.py__call__VszAudio.__call__valuerreturnc Cszddl}Wnty}ztd|d}~ww|dur tdtjr*ddlm}nd}t|tr6d|dSt|t t frB|ddS|durOt||rOt |Sd|vrkt }|j ||dj|dd d |ddS|d durtj|d r|d d r|ddurtd |drtj|dtjdtjd}ntj|d dddtjd}t d}|j |||dd d |ddSd|d dS|ddus|d dur|d|d dStd|d)zEncode example into a format for Arrow. Args: value (`str`, `bytes`,`bytearray`,`dict`, `AudioDecoder`): Data passed as input to Audio feature. Returns: `dict` rN;To support encoding audio data, please install 'soundfile'.zvalue must be providedrrarrayrwavformatr!pcmzBTo use PCM files, please specify a 'sampling_rate' in Audio objectr )rihr)rmodezUAn audio sample should have one of 'path' or 'bytes' but they are missing or None in .) soundfile ImportError ValueErrorr TORCHCODEC_AVAILABLEtorchcodec.decodersr isinstancestrr bytearrayencode_torchcodec_audiorwriteTgetvaluegetosr!isfileendswithKeyErrornp frombufferint16astypefloat32memmap)r&r*sferrrbuffer bytes_valuer'r'r(encode_exampleYsJ      "  zAudio.encode_exampletoken_per_repo_idc CsJtjr ddlm}ntd|jstd|ddur#|d|dfn|ddf\}}|dur;|dur;td|d |durMt|rM|||j |j d }nL|dur|pTi}| d d }| tj retjntj}t||} | durx|| d nd} t| d} t|d| d} || |j |j d }n |||j |j d }||d|_||j_|S)a,Decode example audio file into audio data. Args: value (`dict`): A dictionary with keys: - `path`: String with relative audio file path. - `bytes`: Bytes of the audio file. token_per_repo_id (`dict`, *optional*): To access and decode audio files from private repositories on the Hub, you can pass a dictionary repo_id (`str`) -> token (`bool` or `str`) Returns: `torchcodec.decoders.AudioDecoder` rrz$z&Audio.cast_storage..r)patypes is_stringrsr-lenrm StructArray from_arraysis_null is_binaryrn is_structget_all_field_indices to_pylistget_field_indexrr r")r&rq bytes_array path_arrayr'r'r( cast_storages"     zAudio.cast_storagecsduritfddtjfdd|Dtd}tjdd|dDtd}tjj||gd dg| d }t ||j S) a8Embed audio files into the Arrow array. Args: storage (`pa.StructArray`): PyArrow array to embed. Returns: `pa.StructArray`: Array in the Audio arrow storage type, that is `pa.struct({"bytes": pa.binary(), "path": pa.string()})`. Ncs|dd}|tjrtjntj}t||}|dur#|dnd}t|d}t |d|d }| WdS1s@wYdS)NrUrVrWrXrZr[) r_r`r rarbrcrrCr rread)r!rgrhrirYr\rj)rSr'r( path_to_bytes s  $z*Audio.embed_storage..path_to_bytescs8g|]}|dur|ddur|dn|dndqS)Nr r!r'rv)rr'r(rys*z'Audio.embed_storage..rrcSs$g|]}|durtj|ndqSr%)rDr!basename)rwr!r'r'r(ryrzr!r rt) rr{r-rrmrrnrrrr r")r&rqrSrrr')rrSr( embed_storages    zAudio.embed_storager%)#__name__ __module__ __qualname____doc__rrint__annotations__rboolrrrr=rrr{structrmrnr"rr$r)r r r>rrRrkrp StringArrayrrrr'r'r'r(rs, 5 $ > 5 &rrfrr+c Cst|dr|jSzddl}Wnty}ztd|d}~ww|}|j}t}|j ||j |j dd| ddS)Nrdrr,r.r/r) hasattrrdr7r8get_all_samplesdatacpunumpyrr@rArTrB)rfrNrOsamplesr-rPr'r'r(r?%s   r?)#rD dataclassesrriortypingrrrrr rrHpyarrowr{r download.download_configr tabler utils.file_utilsrrutils.py_utilsrrr;rrorrrr?r'r'r'r(s&