neuralset.events.etypes.Audio¶

class neuralset.events.etypes.Audio(*, start: float, timeline: str, duration: Annotated[float, Ge(ge=0)] = 0.0, extra: dict[str, Any] = {}, filepath: Path | str = '', frequency: float = 0, offset: Annotated[float, Ge(ge=0)] = 0.0)[source][source]¶

Audio event corresponding to a WAV file.

Requires soundfile to be installed. Duration and frequency are auto-detected from the file if not provided.

Returns:: Audio tensor of shape (num_samples, num_channels) when read() is called
Return type:: torch.Tensor

Notes

Automatically detects frequency and duration from file metadata
Supports partial loading via offset and duration
Mono audio is automatically reshaped to (N, 1)

Examples

audio = Audio(start=0, timeline="audio_exp", filepath="speech.wav")
audio_tensor = audio.read()  # Returns torch.Tensor
print(audio_tensor.shape)  # (num_samples, num_channels)