目录
目录

torchaudio.backend

Overview

torchaudio.backend module provides implementations for audio file I/O functionalities, which are torchaudio.info, torchaudio.load, and torchaudio.save.

There are currently four implementations available.

Note

Instead of calling functions in torchaudio.backend directly, please use torchaudio.info, torchaudio.load, and torchaudio.save with proper backend set with torchaudio.set_audio_backend().

Availability

"sox_io" backend requires C++ extension module, which is included in Linux/macOS binary distributions. This backend is not available on Windows.

"soundfile" backend requires SoundFile. Please refer to the SoundFile documentation for the installation.

Common Data Structure

Structures used to report the metadata of audio files.

AudioMetaData

class torchaudio.backend.common.AudioMetaData(sample_rate: int, num_frames: int, num_channels: int, bits_per_sample: int, encoding: str)[source]

Return type of torchaudio.info function.

This class is used by “sox_io” backend and “soundfile” backend with the new interface.

Variables

  • sample_rate (int) – Sample rate

  • num_frames (int) – The number of frames

  • num_channels (int) – The number of channels

  • bits_per_sample (int) – The number of bits per sample. This is 0 for lossy formats, or when it cannot be accurately inferred.

  • encoding (str) –

    Audio encoding The values encoding can take are one of the following:

    • PCM_S: Signed integer linear PCM

    • PCM_U: Unsigned integer linear PCM

    • PCM_F: Floating point linear PCM

    • FLAC: Flac, Free Lossless Audio Codec

    • ULAW: Mu-law

    • ALAW: A-law

    • MP3 : MP3, MPEG-1 Audio Layer III

    • VORBIS: OGG Vorbis

    • AMR_WB: Adaptive Multi-Rate

    • AMR_NB: Adaptive Multi-Rate Wideband

    • OPUS: Opus

    • HTK: Single channel 16-bit PCM

    • UNKNOWN : None of above

Tutorials using AudioMetaData:

Sox IO Backend

The sox_io backend is available and default on Linux/macOS and not available on Windows.

I/O functions of this backend support TorchScript.

You can switch from another backend to the sox_io backend with the following;

torchaudio.set_audio_backend("sox_io")

info

torchaudio.backend.sox_io_backend.info(filepath: str, format: Optional[str] = None) → torchaudio.backend.common.AudioMetaData[source]

Get signal information of an audio file.

Parameters

  • filepath (path-like object or file-like object) –

    Source of audio data. When the function is not compiled by TorchScript, (e.g. torch.jit.script), the following types are accepted;

    • path-like: file path

    • file-like: Object with read(size: int) -> bytes method, which returns byte string of at most size length.

    When the function is compiled by TorchScript, only str type is allowed.

    Note

    • When the input type is file-like object, this function cannot get the correct length (num_samples) for certain formats, such as mp3 and vorbis. In this case, the value of num_samples is 0.

    • This argument is intentionally annotated as str only due to TorchScript compiler compatibility.

  • format (str or None, optional) – Override the format detection with the given format. Providing the argument might help when libsox can not infer the format from header or extension.

Returns

Metadata of the given audio.

Return type

AudioMetaData

load

torchaudio.backend.sox_io_backend.load(filepath: str, frame_offset: int = 0, num_frames: int = - 1, normalize: bool = True, channels_first: bool = True, format: Optional[str] = None) → Tuple[torch.Tensor, int][source]

Load audio data from file.

Note

This function can handle all the codecs that underlying libsox can handle, however it is tested on the following formats;

  • WAV, AMB

    • 32-bit floating-point

    • 32-bit signed integer

    • 24-bit signed integer

    • 16-bit signed integer

    • 8-bit unsigned integer (WAV only)

  • MP3

  • FLAC

  • OGG/VORBIS

  • OPUS

  • SPHERE

  • AMR-NB

To load MP3, FLAC, OGG/VORBIS, OPUS and other codecs libsox does not handle natively, your installation of torchaudio has to be linked to libsox and corresponding codec libraries such as libmad or libmp3lame etc.

By default (normalize=True, channels_first=True), this function returns Tensor with float32 dtype and the shape of [channel, time]. The samples are normalized to fit in the range of [-1.0, 1.0].

When the input format is WAV with integer type, such as 32-bit signed integer, 16-bit signed integer, 24-bit signed integer, and 8-bit unsigned integer, by providing normalize=False, this function can return integer Tensor, where the samples are expressed within the whole range of the corresponding dtype, that is, int32 tensor for 32-bit signed PCM, int16 for 16-bit signed PCM and uint8 for 8-bit unsigned PCM. Since torch does not support int24 dtype, 24-bit signed PCM are converted to int32 tensors.

normalize parameter has no effect on 32-bit floating-point WAV and other formats, such as flac and mp3. For these formats, this function always returns float32 Tensor with values normalized to [-1.0, 1.0].

Parameters

  • filepath (path-like object or file-like object) –

    Source of audio data. When the function is not compiled by TorchScript, (e.g. torch.jit.script), the following types are accepted;

    • path-like: file path

    • file-like: Object with read(size: int) -> bytes method, which returns byte string of at most size length.

    When the function is compiled by TorchScript, only str type is allowed.

    Note: This argument is intentionally annotated as str only due to TorchScript compiler compatibility.

  • frame_offset (int) – Number of frames to skip before start reading data.

  • num_frames (int, optional) – Maximum number of frames to read. -1 reads all the remaining samples, starting from frame_offset. This function may return the less number of frames if there is not enough frames in the given file.

  • normalize (bool, optional) – When True, this function always return float32, and sample values are normalized to [-1.0, 1.0]. If input file is integer WAV, giving False will change the resulting Tensor type to integer type. This argument has no effect for formats other than integer WAV type.

  • channels_first (bool, optional) – When True, the returned Tensor has dimension [channel, time]. Otherwise, the returned Tensor’s dimension is [time, channel].

  • format (str or None, optional) – Override the format detection with the given format. Providing the argument might help when libsox can not infer the format from header or extension.

Returns

Resulting Tensor and sample rate.

If the input file has integer wav format and normalization is off, then it has integer type, else float32 type. If channels_first=True, it has [channel, time] else [time, channel].

Return type

(torch.Tensor, int)

save

torchaudio.backend.sox_io_backend.save(filepath: str, src: torch.Tensor, sample_rate: int, channels_first: bool = True, compression: Optional[float] = None, format: Optional[str] = None, encoding: Optional[str] = None, bits_per_sample: Optional[int] = None)[source]

Save audio data to file.

Parameters

  • filepath (str or pathlib.Path) – Path to save file. This function also handles pathlib.Path objects, but is annotated as str for TorchScript compiler compatibility.

  • src (torch.Tensor) – Audio data to save. must be 2D tensor.

  • sample_rate (int) – sampling rate

  • channels_first (bool, optional) – If True, the given tensor is interpreted as [channel, time], otherwise [time, channel].

  • compression (float or None, optional) –

    Used for formats other than WAV. This corresponds to -C option of sox command.

    "mp3"

    Either bitrate (in kbps) with quality factor, such as 128.2, or VBR encoding with quality factor such as -4.2. Default: -4.5.

    "flac"

    Whole number from 0 to 8. 8 is default and highest compression.

    "ogg", "vorbis"

    Number from -1 to 10; -1 is the highest compression and lowest quality. Default: 3.

    See the detail at http://sox.sourceforge.net/soxformat.html.

  • format (str or None, optional) –

    Override the audio format. When filepath argument is path-like object, audio format is infered from file extension. If file extension is missing or different, you can specify the correct format with this argument.

    When filepath argument is file-like object, this argument is required.

    Valid values are "wav", "mp3", "ogg", "vorbis", "amr-nb", "amb", "flac", "sph", "gsm", and "htk".

  • encoding (str or None, optional) –

    Changes the encoding for the supported formats. This argument is effective only for supported formats, such as "wav", ""amb" and "sph". Valid values are;

    • "PCM_S" (signed integer Linear PCM)

    • "PCM_U" (unsigned integer Linear PCM)

    • "PCM_F" (floating point PCM)

    • "ULAW" (mu-law)

    • "ALAW" (a-law)

    Default values

    If not provided, the default value is picked based on format and bits_per_sample.

    "wav", "amb"
    • If both encoding and bits_per_sample are not provided, the dtype of the
      Tensor is used to determine the default value.

      • "PCM_U" if dtype is uint8

      • "PCM_S" if dtype is int16 or int32

      • "PCM_F" if dtype is float32

    • "PCM_U" if bits_per_sample=8

    • "PCM_S" otherwise

    "sph" format;

    • the default value is "PCM_S"

  • bits_per_sample (int or None, optional) –

    Changes the bit depth for the supported formats. When format is one of "wav", "flac", "sph", or "amb", you can change the bit depth. Valid values are 8, 16, 32 and 64.

    Default Value;

    If not provided, the default values are picked based on format and "encoding";

    "wav", "amb";
    • If both encoding and bits_per_sample are not provided, the dtype of the
      Tensor is used.

      • 8 if dtype is uint8

      • 16 if dtype is int16

      • 32 if dtype is int32 or float32

    • 8 if encoding is "PCM_U", "ULAW" or "ALAW"

    • 16 if encoding is "PCM_S"

    • 32 if encoding is "PCM_F"

    "flac" format;

    • the default value is 24

    "sph" format;

    • 16 if encoding is "PCM_U", "PCM_S", "PCM_F" or not provided.

    • 8 if encoding is "ULAW" or "ALAW"

    "amb" format;

    • 8 if encoding is "PCM_U", "ULAW" or "ALAW"

    • 16 if encoding is "PCM_S" or not provided.

    • 32 if encoding is "PCM_F"

Supported formats/encodings/bit depth/compression are;

"wav", "amb"

  • 32-bit floating-point PCM

  • 32-bit signed integer PCM

  • 24-bit signed integer PCM

  • 16-bit signed integer PCM

  • 8-bit unsigned integer PCM

  • 8-bit mu-law

  • 8-bit a-law

Note: Default encoding/bit depth is determined by the dtype of the input Tensor.

"mp3"

Fixed bit rate (such as 128kHz) and variable bit rate compression. Default: VBR with high quality.

"flac"

  • 8-bit

  • 16-bit

  • 24-bit (default)

"ogg", "vorbis"

  • Different quality level. Default: approx. 112kbps

"sph"

  • 8-bit signed integer PCM

  • 16-bit signed integer PCM

  • 24-bit signed integer PCM

  • 32-bit signed integer PCM (default)

  • 8-bit mu-law

  • 8-bit a-law

  • 16-bit a-law

  • 24-bit a-law

  • 32-bit a-law

"amr-nb"

Bitrate ranging from 4.75 kbit/s to 12.2 kbit/s. Default: 4.75 kbit/s

"gsm"

Lossy Speech Compression, CPU intensive.

"htk"

Uses a default single-channel 16-bit PCM format.

Note

To save into formats that libsox does not handle natively, (such as "mp3", "flac", "ogg" and "vorbis"), your installation of torchaudio has to be linked to libsox and corresponding codec libraries such as libmad or libmp3lame etc.

Soundfile Backend

The "soundfile" backend is available when SoundFile is installed. This backend is the default on Windows.

You can switch from another backend to the "soundfile" backend with the following;

torchaudio.set_audio_backend("soundfile")

info

torchaudio.backend.soundfile_backend.info(filepath: str, format: Optional[str] = None) → torchaudio.backend.common.AudioMetaData[source]

Get signal information of an audio file.

Note

filepath argument is intentionally annotated as str only, even though it accepts pathlib.Path object as well. This is for the consistency with "sox_io" backend, which has a restriction on type annotation due to TorchScript compiler compatiblity.

Parameters

  • filepath (path-like object or file-like object) – Source of audio data.

  • format (str or None, optional) – Not used. PySoundFile does not accept format hint.

Returns

meta data of the given audio.

Return type

AudioMetaData

load

torchaudio.backend.soundfile_backend.load(filepath: str, frame_offset: int = 0, num_frames: int = - 1, normalize: bool = True, channels_first: bool = True, format: Optional[str] = None) → Tuple[torch.Tensor, int][source]

Load audio data from file.

Note

The formats this function can handle depend on the soundfile installation. This function is tested on the following formats;

  • WAV

    • 32-bit floating-point

    • 32-bit signed integer

    • 16-bit signed integer

    • 8-bit unsigned integer

  • FLAC

  • OGG/VORBIS

  • SPHERE

By default (normalize=True, channels_first=True), this function returns Tensor with float32 dtype and the shape of [channel, time]. The samples are normalized to fit in the range of [-1.0, 1.0].

When the input format is WAV with integer type, such as 32-bit signed integer, 16-bit signed integer and 8-bit unsigned integer (24-bit signed integer is not supported), by providing normalize=False, this function can return integer Tensor, where the samples are expressed within the whole range of the corresponding dtype, that is, int32 tensor for 32-bit signed PCM, int16 for 16-bit signed PCM and uint8 for 8-bit unsigned PCM.

normalize parameter has no effect on 32-bit floating-point WAV and other formats, such as flac and mp3. For these formats, this function always returns float32 Tensor with values normalized to [-1.0, 1.0].

Note

filepath argument is intentionally annotated as str only, even though it accepts pathlib.Path object as well. This is for the consistency with "sox_io" backend, which has a restriction on type annotation due to TorchScript compiler compatiblity.

Parameters

  • filepath (path-like object or file-like object) – Source of audio data.

  • frame_offset (int, optional) – Number of frames to skip before start reading data.

  • num_frames (int, optional) – Maximum number of frames to read. -1 reads all the remaining samples, starting from frame_offset. This function may return the less number of frames if there is not enough frames in the given file.

  • normalize (bool, optional) – When True, this function always return float32, and sample values are normalized to [-1.0, 1.0]. If input file is integer WAV, giving False will change the resulting Tensor type to integer type. This argument has no effect for formats other than integer WAV type.

  • channels_first (bool, optional) – When True, the returned Tensor has dimension [channel, time]. Otherwise, the returned Tensor’s dimension is [time, channel].

  • format (str or None, optional) – Not used. PySoundFile does not accept format hint.

Returns

Resulting Tensor and sample rate.

If the input file has integer wav format and normalization is off, then it has integer type, else float32 type. If channels_first=True, it has [channel, time] else [time, channel].

Return type

(torch.Tensor, int)

save

torchaudio.backend.soundfile_backend.save(filepath: str, src: torch.Tensor, sample_rate: int, channels_first: bool = True, compression: Optional[float] = None, format: Optional[str] = None, encoding: Optional[str] = None, bits_per_sample: Optional[int] = None)[source]

Save audio data to file.

Note

The formats this function can handle depend on the soundfile installation. This function is tested on the following formats;

  • WAV

    • 32-bit floating-point

    • 32-bit signed integer

    • 16-bit signed integer

    • 8-bit unsigned integer

  • FLAC

  • OGG/VORBIS

  • SPHERE

Note

filepath argument is intentionally annotated as str only, even though it accepts pathlib.Path object as well. This is for the consistency with "sox_io" backend, which has a restriction on type annotation due to TorchScript compiler compatiblity.

Parameters

  • filepath (str or pathlib.Path) – Path to audio file.

  • src (torch.Tensor) – Audio data to save. must be 2D tensor.

  • sample_rate (int) – sampling rate

  • channels_first (bool, optional) – If True, the given tensor is interpreted as [channel, time], otherwise [time, channel].

  • compression (float of None, optional) – Not used. It is here only for interface compatibility reson with “sox_io” backend.

  • format (str or None, optional) –

    Override the audio format. When filepath argument is path-like object, audio format is inferred from file extension. If the file extension is missing or different, you can specify the correct format with this argument.

    When filepath argument is file-like object, this argument is required.

    Valid values are "wav", "ogg", "vorbis", "flac" and "sph".

  • encoding (str or None, optional) –

    Changes the encoding for supported formats. This argument is effective only for supported formats, sush as "wav", ""flac" and "sph". Valid values are;

    • "PCM_S" (signed integer Linear PCM)

    • "PCM_U" (unsigned integer Linear PCM)

    • "PCM_F" (floating point PCM)

    • "ULAW" (mu-law)

    • "ALAW" (a-law)

  • bits_per_sample (int or None, optional) – Changes the bit depth for the supported formats. When format is one of "wav", "flac" or "sph", you can change the bit depth. Valid values are 8, 16, 24, 32 and 64.

Supported formats/encodings/bit depth/compression are:

"wav"

  • 32-bit floating-point PCM

  • 32-bit signed integer PCM

  • 24-bit signed integer PCM

  • 16-bit signed integer PCM

  • 8-bit unsigned integer PCM

  • 8-bit mu-law

  • 8-bit a-law

Note:

Default encoding/bit depth is determined by the dtype of the input Tensor.

"flac"

  • 8-bit

  • 16-bit (default)

  • 24-bit

"ogg", "vorbis"

  • Doesn’t accept changing configuration.

"sph"

  • 8-bit signed integer PCM

  • 16-bit signed integer PCM

  • 24-bit signed integer PCM

  • 32-bit signed integer PCM (default)

  • 8-bit mu-law

  • 8-bit a-law

  • 16-bit a-law

  • 24-bit a-law

  • 32-bit a-law

文档

访问 PyTorch 的全面开发人员文档

查看文档

教程

获取面向初学者和高级开发人员的深入教程

查看教程

资源

查找开发资源并解答您的问题

查看资源