mariaKlis/MLMoodLamp

Dataset Card for Dataset Name

Welcome to my emotion voice recognition dataset in Edge Impulse, which is made for Mood Lamp project. It includes 4 labels: depressed, agitated, neutral and noise. Each recording lasts 3s. The dataset is collected by 6 people and 2 AI voices.

This dataset card aims to be a base template for new datasets. It has been generated using this raw template.

Dataset Details

Dataset Description

This dataset was created for the Mood Lamp project, an emotion-recognition system based on short voice recordings. It contains 3-second English audio clips of speakers saying “Today, I feel like this,” labelled into four classes: depressed, agitated, neutral, and noise. The recordings come from six human speakers and two AI voices, providing variation in tone and expression. Curated by: Maria Klis Funded by: Not funded Shared by: Maria Klis Language(s) (NLP): English License: Creative Commons Attribution–NonCommercial 4.0 (CC BY-NC 4.0)

Uses

This dataset supports the Mood Lamp project, which uses voice-based emotion recognition to offer gentle, non-clinical emotional support. The model classifies short voice samples and triggers ambient light responses—yellow for depressed, green for agitated, and white for neutral—based on calm-technology principles. The goal is not diagnosis but creating a soothing, low-threshold tool for people experiencing stress, depression, or anxiety.

Direct Use

The dataset is suitable for research and development of small-scale emotion recognition systems. Ideal use cases include:

• Training models to detect emotional states from short voice recordings.

• Exploring non-clinical emotional support tools like the Mood Lamp.

• Developing ambient feedback systems that respond to emotions with light or other cues.

• Studying applications of affective computing and calm-technology principles.

It is not intended for clinical diagnosis or high-stakes decision-making.

Out-of-Scope Use

This dataset is not meant for medical or clinical use and works best with short, controlled recordings. It may not perform well on regular conversations or other languages. Use it only for research or non-clinical emotional support projects.

Dataset Structure

The dataset contains 3-second English audio recordings of “Today, I feel like this”, labeled as depressed, agitated, neutral, or noise. Recordings come from six humans and two AI voices. Files are stored in a flat structure with labels in filenames or a metadata file.

Dataset Creation

Curation Rationale

This dataset was created for the Mood Lamp project to explore how voice-based emotion recognition can provide gentle, non-clinical emotional support. By collecting short, labeled voice recordings, the dataset enables training models that map emotions to ambient light responses, helping users manage mood through subtle, calming cues.

Source Data

The dataset contains original English recordings of “Today, I feel like this”, captured on a phone microphone by six humans and two AI voices. No external sources were used.

Data Collection and Processing

Recordings were collected using a phone microphone and Edge Impulse, with each sample set to 3 seconds. Files were manually labeled.

Who are the source data producers?

The dataset was created by six human speakers: three females and three males, from Denmark, Poland, Italy, and Thailand. In addition, two AI-generated voices from Eleven Labs were used.

Annotation process

Labels and recording length were set manually in the Edge Impulse interface. No additional annotation tools or formal validation were used.

Who are the annotators?

The annotations were created by the dataset creator, Maria Klis, during data collection in Edge Impulse.

Personal and Sensitive Information

The dataset does not contain personal, sensitive, or private information. All recordings consist of a single sentence (“Today, I feel like this”). All human participants gave their consent for the recordings to be used in this dataset.

Bias, Risks, and Limitations

Only short, 3-second recordings of one sentence; small, limited speaker set. Not for clinical use, only for research or non-clinical emotional support.

Recommendations

Use this dataset only for research or non-clinical projects. It may not work well on other sentences or languages. Adding more voices can help reduce bias and improve results.

BibTeX: @misc{klis_moodlamp_2025, author = {Maria Klis}, title = {Mood Lamp Voice Emotion Recognition Dataset}, year = {2025}, howpublished = {\url{https://studio.edgeimpulse.com/public/829373/live}}, note = {3-second English audio recordings labeled as depressed, agitated, neutral, or noise. CC BY-NC 4.0} }

APA: Klis, M. (2025). Mood Lamp Voice Emotion Recognition Dataset [Dataset].Retrieved from https://studio.edgeimpulse.com/public/829373/live

Dataset Card Contact

Mail: mklis25@student.aau.dk