Interspeech2026-Audio-Encoder-Challenge

20-Hour Non-speech Dataset from DataoceanAI

1. Dataset Overview

This dataset is constructed by extracting non-speech data from 8 original datasets, with a total duration of approximately 20 hours. It mainly contains environmental noise in various scenarios, suitable for research fields such as speech signal processing, noise suppression, and acoustic model training.

2. Dataset Composition

2.1 Overall Composition

2.2 Detailed Information of Each Original Dataset

King-ASR-457

King-ASR-610

King-ASR-719

King-ASR-829

King-ASR-862

King-ASR-876

King-ASR-955

King-ASR-958

3. Data Usage

This dataset can be widely used in the following research and application scenarios:

  1. Training and testing of speech noise suppression algorithms
  2. Construction of acoustic environment classification models
  3. Optimization of anti-noise performance of speech recognition systems
  4. Evaluation of audio signal processing algorithms
  5. Improvement of environmental adaptability of human-computer interaction systems

4. Data Download Method

4.1 Download Process

  1. Application Registration: Users who wish to obtain the dataset need to register first
  2. Eligibility Review: Staff will review the registration information to confirm compliance with usage conditions
  3. Obtain Link: After passing the review, staff will send the data download link via private message

4.2 Notes

5. Version Information


The final interpretation right of this description document belongs to the data provider