mltk.datasets.audio.mit_ir_survey

MIT Impulse Response Survey

This is a dataset of environmental impulse responses from various real-world locations, collected in the MIT IR Survey by Traer and McDermott.

Each audio file is a waveform which contains the impulse response of a location. That is to say, how an instantaneous pressure at $t = 0$ is reflected, damped and scattered in the environment.

By convolving the dataset impulse responses with an audio clip, we can simulate how that audio clip would sound if emitted and recorded in the environment where the impulse response was recorded. This is a technique commonly used for data augmentation in audio processing problems, commonly referred to as multi-style training (see Deep Spoken Keyword Spotting: An Overview), simulated reverberation (see e.g. End-to-End Streaming Keyword Spotting) or acoustic simulation.

License

CC-BY 4.0 (see MIT Creative Commons License for details).

Credits

Traer and McDermott 2016 paper Statistics of natural reverberation enable perceptual separation of sound and space

Variables

DOWNLOAD_URL	Public download URL
VERIFY_SHA1	SHA1 hash of the downloaded archive file

Functions

apply_ir(audio, ir)	Apply an impulse response to the given audio sample
apply_random_ir(audio, ir_samples[, seed])	Appyly a random impulse response to the given audio sample
download([dest_dir, dest_subdir, logger, ...])	Download and extract the dataset
load_dataset(dataset_dir)	Load the impulse response dataset directory into RAM

DOWNLOAD_URL = 'https://mcdermottlab.mit.edu/Reverb/IRMAudio/Audio.zip'

Public download URL

VERIFY_SHA1 = 'de04f5be419c12f4f847f65d7ef8e2356b73aa38'

SHA1 hash of the downloaded archive file

download(dest_dir=None, dest_subdir='datasets/mit_ir_survey', logger=None, clean_dest_dir=False, sample_rate_hz=16000)

Download and extract the dataset

Return type:

str

Returns:

The directory path to the extracted dataset

Parameters:

• dest_dir (str) –

• logger (Logger) –

apply_ir(audio, ir)

Apply an impulse response to the given audio sample

Return type:

ndarray

Parameters:

• audio (ndarray) –

• ir (ndarray) –

load_dataset(dataset_dir)

Load the impulse response dataset directory into RAM

Return type:

List[ndarray]

Parameters:

dataset_dir (str) –

apply_random_ir(audio, ir_samples, seed=42)

Appyly a random impulse response to the given audio sample

Return type:

ndarray

Parameters:

• audio (ndarray) –

• ir_samples (List[ndarray]) –

• seed (int) –