Skip to content
Karmyug

Wake words and voice commands

Short command sets recorded at varied distance, angle and background noise.

25 languagesConsumer devices

Overview

Wake word and short command capture designed for on-device models. Each speaker records at multiple distances and angles, in rooms with controlled noise profiles, so the set covers the far-field conditions where these models actually fail.

Specification

Format
WAV, 16 kHz
Distance
0.3 m to 5 m
Angles
0, 45, 90 degrees
Environments
Quiet, TV, kitchen, street

Sample

Illustrative sample. It shows the structure, fields and metadata of this dataset with invented values, not real recordings or records. A real sample pack is sent on request.

Sample records for Wake words and voice commands
fileduration_slanguagespeaker_idage_bandgenderdeviceenvironmentcontent_noteconsent_ref
sample_0001.wav6.8en-INspk_031018-25femaleAndroid mid-range phoneQuiet roomCaller asks about the status of a delayed paymentconsent_02100
sample_0002.wav11.2en-INspk_031726-35maleAndroid budget phoneStreet, light trafficAgent verifies identity and explains the next stepconsent_02101
sample_0003.wav8.4en-INspk_032436-45femaleiPhoneOffice, background talkSpeaker switches language mid-sentence to name a productconsent_02102
sample_0004.wav14.9en-INspk_033146-60maleAndroid mid-range phoneVehicle, windows upCustomer reports a network problem at homeconsent_02103
sample_0005.wav3.1en-INspk_033826-35femaleLaptop microphoneHome, TV onShort voice command with a number in itconsent_02104

Quality assurance

  • Automated duration and VAD check
  • Distance verification
  • Negative sample balancing

Want a real sample before you commit?

The sample above shows the structure. We also send a representative slice that includes the hard cases, plus the QA report and licence terms. No call required.