Soundscapes in Spectrograms: Pioneering Multilabel Classification for South Asian Sounds

Sudip Chakrabarty; Pappu Bishwas; Rajdeep Chatterjee; Tathagata Bandyopadhyay; Digonto Biswas; Bibek Howlader

cs.SDMar 9, 2026

Soundscapes in Spectrograms: Pioneering Multilabel Classification for South Asian Sounds

Authors: Sudip Chakrabarty, Pappu Bishwas, Rajdeep Chatterjee, Tathagata Bandyopadhyay, Digonto Biswas, Bibek Howlader

Organizations: School of Computer Engineering, KIIT Deemed to be University, Bhubaneswar, India · School of Computation, Information and Technology, Technical University of Munich, Germany · School of Computer Engineering, American International University, Dhaka, Bangladesh

Abstract

Environmental sound classification is a field of growing importance for urban monitoring and cultural soundscape analysis, especially within the acoustically rich environments of South Asia. These regions present a unique challenge as multiple natural, human, and cultural sounds often overlap, straining traditional methods that frequently rely on Mel Frequency Cepstral Coefficients (MFCC). This study introduces a novel spectrogram-based methodology with a superior ability to capture these complex auditory patterns. A Convolutional Neural Network (CNN) architecture is implemented to solve a demanding multilabel, multiclass classification problem on the SAS-KIIT dataset. To demonstrate robustness and comparability, the approach is also validated using the renowned UrbanSound8K dataset. The results confirm that the proposed spectrogram-based method significantly outperforms existing MFCC-based techniques, achieving higher classification accuracy across both datasets. This improvement lays the groundwork for more robust and accurate audio classification systems in real-world applications.

Figures & tables

1. Tanpura	12. Launch Engine
2. Traditional Song	13. Flute
3. Railway Engine	14. Buddhist Prayer
4. Children Class Noise	15. Fish Market
5. Harmonium	16. Tiger
6. Dhak	17. Elephant
7. Tabla	18. Kalboishakhi Storm

Table 1: List of 21 Sound Classes in the SAS-KIIT Dataset

Figure 1: t-SNE: A Representation of Class Distributions in the Feature Space for SAS-KIIT Dataset.

1. Air Conditioner	6. Engine Sound
2. Vehicle Horn	7. Gun Shot
3. Children Playing	8. Jackhammer
4. Dog Bark	9. Siren
5. Drilling	10. Music

Table 2: List of 10 Sound Classes in the UrbanSound8K Dataset

Figure 2: t-SNE: A Representation of Class Distributions in the Feature Space for UrbanSound8K Dataset.

Figure 3: Audio Mixing Process and Label Interactions.

Figure 4: Mixed audio is converted to Mel-spectrograms and fed into a CNN with ReLU and max pooling to extract features; dense layers output multilabel predictions, with dusty pink highlighting detected sound classes in an example recording.

Dataset	Feature	P	R	F1	Acc.
SAS-KIIT	MFCC	0.83	0.73	0.76	93.91
	Spectrogram	0.87	0.78	0.81	95.42
UrbanSound8K	MFCC	0.71	0.61	0.62	84.16
	Spectrogram	0.72	0.67	0.64	86.36

Table 3: Performance comparison of MFCC and Spectrogram features on datasets with fixed no. mixed audio samples.

Dataset	Feature	P	R	F1	Acc.
SAS-KIIT	MFCC	0.84	0.76	0.77	94.63
	Spectrogram	0.89	0.83	0.84	96.37
UrbanSound8K	MFCC	0.68	0.65	0.62	83.94
	Spectrogram	0.69	0.63	0.61	85.26

Table 4: Performance comparison of MFCC and Spectrogram features on datasets with variable no. mixed audio samples.

Figure 5: Log-Scale Representation of Predicted Audio Classes: Orange bars indicate classes detected in the sample audio; Sky-blue bars show other predicted classes with lower probability scores in log scale; The dashed black line (- - -) represents the decision threshold, above which classes are considered present.

Dataset	Model	P	R	F1	Acc.
SAS-KIIT	FACE [ 15 ]	0.91	0.79	0.80	95.22
	PANNs [ 16 ]	0.85	0.76	0.78	92.51
	Proposed Model	0.89	0.83	0.84	96.37
Urban-Sound8K	FACE [ 15 ]	0.71	0.57	0.59	84.54
	PANNs [ 16 ]	0.63	0.61	0.66	83.26
	Proposed Model	0.69	0.63	0.61	85.26

Table 5: Performance comparison of the Proposed Model, FACE and PANNs on datasets using spectrogram features with variable number of mixed audio samples.

Audio Classification

Audio classification assigns recordings to categories such as speech, music, animal calls, acoustic scenes, and health-related events.

See all 147 papers

A Hierarchical Method for Environmental Sounds Recognition in Complex Acoustic Scenes
Oct 26, 2026Zhouwen Li, Haijun Lin, Bowen Lv +1Audio Classification Hierarchical Classification
Hunan Normal University School of Engineering and Design Changsha, China
Music Genre Classification: A Comparative Analysis of Classical Machine Learning and Deep Learning Approaches
Mar 16, 2026Sachin Prajuli, Abhishek Karna, OmPrakash DhaklAudio Classification Music Information Retrieval
Computer Science and Engineering, University of texas at Arlington, Arlington, USA · Electronics and Computer Engineering, Tribhuvan University, IOE, kathmandu · Electronics and Computer Engineering, Tribhuvan University, IOE, Dharan
CoarseSoundNet: Building a reliable model for ecological soundscape analysis
May 20, 2026Alexander Gebhard, Andreas Triantafyllopoulos, Dominik Arend +4Sound Event Detection Passive Acoustic Monitoring
TUM University Hospital, CHI – Chair of Health Informatics, Ismaninger Str. 22, Munich, 81675, Bavaria, Germany · MCML – Munich Center for Machine Learning, Munich, Bavaria, Germany · University of Freiburg, Faculty of Biology, Geobotany, Schaenzlestr. 1, Freiburg, 79104, Baden-Württemberg, Germany +1

Oct 26, 2026J/K move · Enter open · S save

A Hierarchical Method for Environmental Sounds Recognition in Complex Acoustic Scenes

Zhouwen Li, Haijun Lin, Bowen Lv, Yupeng Chen

Hunan Normal University School of Engineering and Design Changsha, China

Audio Classification Hierarchical Classification Passive Acoustic Monitoring

Open paper AskPDF arXiv

Environmental noise identification is essential for noise control, yet conventional single-stage classifiers suffer from class confusion under imbalanced multiclass conditions. This paper proposes a lightweight hierarchical model integrating a Frequency-Time Adaptive Network (FTANet) with a two-stage classification mechanism. FTANet employs parallel asymmetric one-dimensional convolutions to capture temporal dynamics and frequency-band distributions, while adaptive fusion attention (AFA) enhances discriminative acoustic features. In the first stage, FTANet classifies sounds into five coarse-grained categories covering 25 fine-grained classes. In the second stage, samples are routed to the corresponding category-specific classifiers for fine-grained recognition. On a self-constructed dataset, FTANet achieves 95.45% coarse-grained accuracy with only 0.2004M parameters. On UrbanSound8K, it achieves 79.24% mean accuracy and 80.37% mean F1-score in ten-fold evaluation, outperforming compared convolutional models. Field tests on the Tianjin Online Acoustic Environment Monitoring Platform yield 87.94% coarse-grained accuracy and 97.21% conditional fine-grained accuracy for natural sounds correctly routed to C2. The complete hierarchical model achieves 87.18% overall accuracy and 87.40% weighted-average F1-score, demonstrating its feasibility for urban environmental noise monitoring.

Abstract

Audio Classification

Explore similar work

A Hierarchical Method for Environmental Sounds Recognition in Complex Acoustic Scenes

Music Genre Classification: A Comparative Analysis of Classical Machine Learning and Deep Learning Approaches

CoarseSoundNet: Building a reliable model for ecological soundscape analysis