o
z3i: @ s d dl Z d dlmZmZ d dlmZ d dlmZ d dlm Z m
Z
m Z m Z m
Z
d dlZd dlZddlmZ ddlmZ dd lmZ dd
lmZmZ dd lm Z m Z e r^d d l m Z d
dl m!Z! eG dd dZ"ddde#fddZ$dS ) N) dataclassfield)BytesIO)Path)
TYPE_CHECKINGAnyClassVarOptionalUnion )config)DownloadConfig)
array_cast)
is_local_pathxopen)no_op_if_value_is_nullstring_to_dict AudioDecoder ) FeatureTypec @ s^ e Zd ZU dZdZee ed< dZe ed< dZ
ee ed< dZ ee ed< e ddd Z
ee ed
< d Zee ed < ee e d
Zee ed< e d dddZeed< dd Zdeeeee df de fddZ d#de dee eeee df f ddfddZ dede edf f fd d Z d eej ej!f dej!fd d Z"d#d ej!dej!fd!d"Z#dS )$Audioa
Audio [`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 `pathlib.Path`: 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.
num_channels (`int`, *optional*):
The desired number of channels of the samples. By default, the number of channels of the source is used.
Audio decoding will return samples with shape (num_channels, num_samples)
Currently `None` (number of channels of the source, default), `1` (mono) or `2` (stereo) channels are supported.
The `num_channels` argument is passed to `torchcodec.decoders.AudioDecoder`.
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, num_channels=2))
>>> 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 num_channels stream_indexF)defaultrepriddictdtypebytespathpa_type)r initr _typec C s | j S N)r$ )self r) e/lsinfo/ai/hellotax_ai/llm_service/venv_embed/lib/python3.10/site-packages/datasets/features/audio.py__call__] s zAudio.__call__valuer returnc
C s. z ddl }ddlm} W n ty } ztd|d}~ww |du r&tdtjr0ddlm} nd}t |t
r token (`bool` or `str`)
Returns:
`torchcodec.decoders.AudioDecoder`
r r z r RuntimeErrorr? r r r r split
startswith HF_ENDPOINTHUB_DATASETS_URLHUB_DATASETS_HFFS_URLr rM r
r _hf_encodedmetadatar# )
r( r, rY r r# r" audio
source_urlpatternsource_url_fieldsr^ ra fr) r) r* decode_example s< ,
zAudio.decode_exampler c C s, ddl m} | jr
td|d|ddS )z[If in the decodable state, raise an error, otherwise flatten the feature into a dictionary.r )Valuez'Cannot flatten a decoded Audio feature.binarystringr! )featuresrq r r? )r( rq r) r) r* flatten s z
Audio.flattenstoragec C s t j|jr%t jdgt| t d}t jj||gddg| d}nt j
|jrQt |t }t jdgt| t d}t jj||gddg| d}nt j
|jrvt jdgt| t d}t jj||gddg| d}njt j|jr|jdrt dd |jd d
D }nNt j|jr|jdd kr|d}nt jdgt| t d}|jdd kr|d}nt jdgt| t d}t jj||gddg| d}t || jS ) a Cast an Arrow array to the Audio arrow storage type.
The Arrow types that can be converted to the Audio pyarrow storage type are:
- `pa.string()` - it must contain the "path" data
- `pa.binary()` - it must contain the audio bytes
- `pa.struct({"bytes": pa.binary()})`
- `pa.struct({"path": pa.string()})`
- `pa.struct({"bytes": pa.binary(), "path": pa.string()})` - order doesn't matter
Args:
storage (`Union[pa.StringArray, pa.StructArray]`):
PyArrow array to cast.
Returns:
`pa.StructArray`: Array in the Audio arrow storage type, that is
`pa.struct({"bytes": pa.binary(), "path": pa.string()})`
Ntyper" r# maskr1 c S s$ g | ]}|d urt |nd qS r' )r rX .0xr) r) r*
$ z&Audio.cast_storage..F)zero_copy_onlyr )patypes is_stringrx r1 lenrr StructArray from_arraysis_nullis_large_binaryr rs is_binary is_structget_all_field_indicesto_numpyget_field_indexr r$ )r( rv bytes_array
path_arrayr) r) r* cast_storage s2 zAudio.cast_storagec s du ri t fdd tj fdd| D t d}tjdd |d D t d}tjj||gd dg| d
}t
|| j S ) a8 Embed 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()})`.
Nc s | dd }|tjrtjntj}t||}|d ur# |d nd }t|d}t | d|d }|
W d S 1 s@w Y d S )NrZ r[ r\ r] r_ r` ) rd re r rf rg rh r rM r
r read)r# rl rm rn r^ ra ro )rY r) r*
path_to_bytes& s
$z*Audio.embed_storage..path_to_bytesc s8 g | ]}|d ur|d d u r |d n|d nd qS )Nr" r# r) r{ )r r) r* r~ 3 s *z'Audio.embed_storage..rw c S s$ g | ]}|d urt j|nd qS r' )rN r# basename)r| r# r) r) r* r~ : r r# r" ry ) r r r1 to_pylistrr r rs r r r r r$ )r( rv rY r r r) )r rY r*
embed_storage s
zAudio.embed_storager' )$__name__
__module__ __qualname____doc__r r int__annotations__r boolr r r r rC r r r structrr rs r$ r r& r+ r
r" rE r rX rp ru StringArrayr r r r) r) r) r* r s.
: $ E
; .r rk r r- c
C s t | dr| jS zddlm} W n ty! } ztd|d }~ww | }t }|jjd }||j |j
dj |d|d | d dS ) Nri r r. r0 r2 r4 r5 r! )
hasattrri r= r/ r> get_all_samplesr datashapecpur3 rK rL )rk r/ rU samplesrV r r) r) r* rF A s
rF )%rN dataclassesr r ior pathlibr typingr r r r r
numpyrI pyarrowr r download.download_configr
tabler utils.file_utilsr r utils.py_utilsr r rA r rt r r r rF r) r) r) r* s( +