Annotation dimensions
Valence
The horizontal axis runs from negative to positive emotion on a 0–1 scale.
Arousal
The vertical axis runs from low to high emotional energy. Arousal is not the same as loudness.
Perceived emotion is what the audio expresses; felt emotion is what you experience while listening. Select one before starting. It remains fixed for that annotation session.
Research and citation
Yang and colleagues examined how agreement between listeners changes during a live musical performance, then explored reasons for differing perceptions in a laboratory study. The article was first published online in 2021 and appeared in volume 14 in 2023.
Implementation and limits
The original study used time-stamped clicks and, for part of its analysis, step resampling at 1 Hz. This edition adds general audio, local drafts, interval re-recording and 10 Hz derived exports. These are engineering extensions, not a complete replication of the experimental protocol.
Only user inputs are observations. Within a recording interval, each value is held until the next input; gaps remain empty. Raw inputs and edit history are retained. Pointer movement is recorded at up to 30 Hz. Export frequency does not imply measurement precision, and response delay is not corrected.
Each new input records coordinates, audio position and a device-clock UTC timestamp. Sessions retain the interface configuration used during annotation.
Emotion words
The VA plane uses red, orange, blue and green hues. A selection circle with a diameter of one-eighth of the plane width shows up to five nearby words, ordered by distance below the plane. If no words fall within the circle, none are shown.
The 97 words, coordinates and Chinese translations preserve the supplied list. As described in the thesis, editorial ILM tags are mapped to ANEW mean VA values.
Audio and data
The public workspace requires no account. Local audio plays in your browser and is not uploaded. Annotations are saved in this browser and are not submitted to a research service. Clearing browser storage removes drafts; export important work. After a refresh, select local audio again.
Audio collections, emotion words, selection circles, colors and axis labels can be configured separately. Saved sessions retain their configuration when settings change. An optional research collection backend is available for separate deployments and is disabled in this demo. No advertising or third-party analytics are included.
Sample audio
Bundled examples are excerpts from recorded piano and string performances, a live human address and an interview selected without a music bed. Recording sources and licenses are shown in the workspace. These clips are not validated research stimuli or normative emotion examples.