background-audio-r1

This release contains background-noise WAV files normalized for wakeword and
speech-command augmentation.

All WAV files are encoded to format used by the wakewords project:

• WAV container
• PCM signed 16-bit little-endian (pcm_s16le)
• Mono
• 16,000 Hz sample rate

manifest.jsonl contains one JSON object per audio file:

{"audio": "example.wav", "duration_ms": 60000}

Background noise downloaded from Freesounds

The Freesounds files were downloaded from Freesounds.com. All sounds were reviewed for permissive license. Credits listed below.

• street-ambient-low-traffic-2 by bulbastre
• street-berlin-field by balou82
• city-street-under-construction-2 by tode_vole
• keyboard-typing by stu556
• shopping-street-ambience by florianreichelt
• mumbai-market by lenguaverde
• street-atmos-w-footsteps by elzbthep
• colombia_street_market by usbmed_ambiences_sound_library
• rainy-street by iliana22
• spanish-wallas-at-the-center-of-mexican-town by chillaquil
• sea-waves by loredenii
• urban-river-with-birds by garuda1982
• franzosenbecken-01 by venethios
• spring-evening by garuda1982
• city-canal by klankbeeld

Google Speech Commands Background Noise Credits

The files with google_scd_ prefix are background noise from the Google Speech Commands v2 dataset.

These audio files are part of the Speech Commands data set, and covered by the
same Creative Commons BY 4.0 license (see the root directory for more details).

They are a collection of sounds that can be mixed in to simulate background
noise when training machine learning models on audio clips. They were all
either collected personally by Pete Warden in July 2017, or generated in the
case of the pink and white noise.