DCASE Task 6 - Clotho-Moment Extracted Features: M2D-CLAP, LAION-CLAP, BEATs, T5

Extracted features for the Clotho-Moment dataset: for DCASE Task 6: Audio Moment Retrieval from Long Audio (2026) Summary: Models (Audio features): M2D-CLAP, LAION-CLAP, BEATs Models (Text features): M2D-CLAP, LAION-CLAP, T5 Background: For extraction, I used DCASE Task 6 given Train : clotho_moment_train_release.jsonl Val : clotho_moment_val_release.jsonl Test : clotho_moment_test_release.jsonl files to load audio and text files, but the task files don't use all the Clotho-Moment audio files; hence, the extracted features have some files missing from the original Clotho-Moment dataset. The shared features are primarily useful for DCASE Task 6: Audio Moment Retrieval from Long Audio (2026), and all the features of the audio and text files necessary to train the model for this task are present in this repository. Features npz files (respective model) Audio 37,051 Text 44,261 File structure: BEATs |___BEATs_audio |___Amsterdam_0.0_60.0.npz |___ : LAION |___LAION_audio |___Amsterdam_0.0_60.0.npz |___ : |___LAION_text |___qid00000.npz |___ : M2D |___M2D_audio |___Amsterdam_0.0_60.0.npz |___ : |___M2D_text |___qid00000.npz |___ : T5 |___T5_text |___qid00000.npz |___ : Related Code To train or evaluate our model's performance, please use these features along with the instructions and code shared on my GitHub repository. GitHub Repository: AMR-encoder-exploration