This paper proposes a class-incremental learning method for multi-label audio classification. The method enables the system to learn new sound classes incrementally without forgetting previously learned classes. The approach uses an incremental learner that independently learns new classes while preserving knowledge of old classes through two distillation losses: cosine similarity-based and Kullback-Leibler divergence-based. These losses minimize discrepancies in feature representations and outputs between successive learners, ensuring knowledge retention. The experiments are conducted on a dataset with 50 sound classes, starting with 30 base classes and adding 5 classes in each of four incremental phases. After each phase, the system is tested on all classes learned so far. The proposed method achieves an average F1-score of 40.9% across five phases, ranging from 45.2% on 30 classes to 36.3% on 50 classes. The average performance degradation is only 0.7 percentage points from the initial F1-score of 45.2%. The paper introduces a novel approach combining independent learning (IndL) with two distillation losses: output discrepancy loss (L^OD) and feature discrepancy loss (L^FD). These losses help preserve knowledge of old classes by minimizing discrepancies in outputs and features between current and previous learners. The proposed method, called IODFD, outperforms other approaches in terms of performance on both old and new classes, achieving an average F1-score of 40.9%, mAP of 25.3%, and the lowest forgetting rate of 0.7 percentage points. The results show that IODFD is both plastic enough to learn new sound classes and stable enough to retain knowledge of old classes. The method is evaluated on a dataset with 50 sound classes, with the initial phase containing 30 base classes and four incremental phases adding 5 classes each. The system is tested after each phase on all classes learned so far, demonstrating its effectiveness in multi-label audio classification. The proposed approach is compared with other methods, including fine-tuning (FT), feature extraction (FE), IndL, L^OD, IOD, and IFD, showing that IODFD achieves the best performance. The method is also compared with AT, a non-incremental approach, and shows that IODFD is competitive with AT in performance. The results indicate that IODFD is effective in handling imbalanced sequential sound classification tasks.This paper proposes a class-incremental learning method for multi-label audio classification. The method enables the system to learn new sound classes incrementally without forgetting previously learned classes. The approach uses an incremental learner that independently learns new classes while preserving knowledge of old classes through two distillation losses: cosine similarity-based and Kullback-Leibler divergence-based. These losses minimize discrepancies in feature representations and outputs between successive learners, ensuring knowledge retention. The experiments are conducted on a dataset with 50 sound classes, starting with 30 base classes and adding 5 classes in each of four incremental phases. After each phase, the system is tested on all classes learned so far. The proposed method achieves an average F1-score of 40.9% across five phases, ranging from 45.2% on 30 classes to 36.3% on 50 classes. The average performance degradation is only 0.7 percentage points from the initial F1-score of 45.2%. The paper introduces a novel approach combining independent learning (IndL) with two distillation losses: output discrepancy loss (L^OD) and feature discrepancy loss (L^FD). These losses help preserve knowledge of old classes by minimizing discrepancies in outputs and features between current and previous learners. The proposed method, called IODFD, outperforms other approaches in terms of performance on both old and new classes, achieving an average F1-score of 40.9%, mAP of 25.3%, and the lowest forgetting rate of 0.7 percentage points. The results show that IODFD is both plastic enough to learn new sound classes and stable enough to retain knowledge of old classes. The method is evaluated on a dataset with 50 sound classes, with the initial phase containing 30 base classes and four incremental phases adding 5 classes each. The system is tested after each phase on all classes learned so far, demonstrating its effectiveness in multi-label audio classification. The proposed approach is compared with other methods, including fine-tuning (FT), feature extraction (FE), IndL, L^OD, IOD, and IFD, showing that IODFD achieves the best performance. The method is also compared with AT, a non-incremental approach, and shows that IODFD is competitive with AT in performance. The results indicate that IODFD is effective in handling imbalanced sequential sound classification tasks.