Google Speech Commands: Keyword Spotting Dataset

# Speech Commands - Keyword Recognition
# Based on Warden, arXiv 2018
# Paper: https://arxiv.org/abs/1804.03209
# Dataset: https://www.tensorflow.org/datasets/catalog/speech_commands
#
# This task asks annotators to listen to short audio clips of spoken commands
# and identify the keyword being spoken. They also assess the audio quality
# to help filter training data.
#
# Command Keywords:
# - Yes, No, Up, Down, Left, Right, On, Off, Stop, Go
# - Unknown: Not one of the target keywords
# - Silence: No speech detected
#
# Audio Quality:
# - Clear: Audio is clean with clearly audible speech
# - Noisy: Background noise present but speech is still identifiable
# - Ambiguous: Speech is unclear or could be multiple words
#
# Annotation Guidelines:
# 1. Listen to the audio clip (replay as needed)
# 2. Label the audio with the spoken command keyword
# 3. Assess the audio quality (Clear, Noisy, or Ambiguous)
# 4. Compare with the expected command shown for reference

annotation_task_name: "Speech Commands - Keyword Recognition"
task_dir: "."

data_files:
  - sample-data.json

item_properties:
  id_key: "id"
  text_key: "text"

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

port: 8000
server_name: localhost

annotation_schemes:
  - annotation_type: radio
    name: audio_quality
    description: "Assess the quality of the audio clip"
    labels:
      - "Clear"
      - "Noisy"
      - "Ambiguous"
    keyboard_shortcuts:
      "Clear": "1"
      "Noisy": "2"
      "Ambiguous": "3"
    tooltips:
      "Clear": "Audio is clean with clearly audible speech"
      "Noisy": "Background noise is present but speech is still identifiable"
      "Ambiguous": "Speech is unclear or could be interpreted as multiple different words"

  - annotation_type: audio_annotation
    name: command_label
    description: "Label the spoken command in the audio clip"
    mode: "label"
    labels:
      - "Yes"
      - "No"
      - "Up"
      - "Down"
      - "Left"
      - "Right"
      - "On"
      - "Off"
      - "Stop"
      - "Go"
      - "Unknown"
      - "Silence"

annotation_instructions: |
  You will be shown an audio clip containing a short spoken command.
  1. Listen to the audio clip (you may replay it as needed).
  2. Label the audio with the command keyword you hear.
     Use "Unknown" if the word is not one of the target keywords.
     Use "Silence" if no speech is detected.
  3. Assess the audio quality: Clear, Noisy, or Ambiguous.

  The expected command is shown for reference, but your label should
  reflect what you actually hear in the audio.

html_layout: |
  <div style="padding: 15px; max-width: 800px; margin: auto;">
    <div style="background: #f0f9ff; border: 1px solid #bae6fd; border-radius: 8px; padding: 16px; margin-bottom: 16px; text-align: center;">
      <audio controls style="width: 100%;">
        <source src="{{audio_url}}" type="audio/wav">
        Your browser does not support the audio element.
      </audio>
    </div>
    <div style="background: #fef3c7; border: 1px solid #fde68a; border-radius: 8px; padding: 12px;">
      <strong style="color: #92400e;">Expected Command:</strong>
      <span style="font-size: 16px; color: #78350f;">{{text}}</span>
    </div>
  </div>

allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 2
allow_skip: true
skip_reason_required: false