spdl.io.get_audio_filter_desc

get_audio_filter_desc(*, sample_rate: int | None = None, num_channels: int | None = None, sample_fmt: str | None = 'fltp', timestamp: tuple[float, float] | None = None, num_frames: int | None = None, filter_desc: str | None = None) → str | None[source]

Construct FFmpeg filter expression for preprocessing audio.

Parameters:

  • sample_rate (int) – Optional: Change the sample rate.

  • num_channels (int) – Optional: Change the number of channels.

  • sample_fmt (str) –

    Optional: Change the format of sample. Valid values are ("u8", "u8p", "s16", "s16p", "s32", "s32p", "flt", "fltp", "s64", "s64p", "dbl", "dblp").

    The suffix "p" means planar format, i.e. when data are converted to Tensor, the shape is (num_channels, num_frames) instead of (num_frames, num_channels).

    Default “fltp”. The audio samples are converted to 32-bit floating point in [-1, 1] range.

  • timestamp (tuple[float, float]) – Optional: Trim the audio by start and end time. This has to match the value passed to demux functions, which can be retrieved from spdl.io.AudioPackets.timestamp.

  • num_frames (int) – Optional: Fix the number of output frames by dropping the exceeding frames or padding with silence.

Returns:

Filter description.