VoIP Quality Prediction: My Experience at UBMK 2022

My Experience at UBMK 2022: Presenting "Quality of Experience Prediction for VoIP Calls Using Audio MFCCs and Multilayer Perceptron"

I recently had the incredible opportunity to attend the UBMK 2022 conference where I had the privilege of presenting my research paper [1] titled "Quality of Experience Prediction for VoIP Calls Using Audio MFCCs and Multilayer Perceptron." This conference provided a platform for researchers and professionals to share their insights and advancements in various fields, and I was excited to contribute to the discussion.

Overview of the Paper

My paper addresses the challenges faced by VoIP service providers in ensuring high-quality communication services for users. VoIP (Voice Over Internet Protocol) technology relies on real-time transmission of voice packets over the internet, making it susceptible to various factors such as packet loss, latency, jitter, codec issues, and encryption.

To maintain a high level of service quality, providers often use Quality of Service (QoS) reports to analyze parameters like packet loss, delay, jitter, and codec information extracted from VoIP calls. However, in some cases, these reports may be insufficient or corrupted. To address this, I proposed a machine learning-based model that can classify packet loss into six classes, a crucial factor affecting the quality of experience for users.

Methodology and Key Findings

The model was trained using 9000 5-second audio recordings from 15 different speakers, represented with Mel Frequency Cepstrum Coefficients (MFCCs). The results were promising, with the model achieving an accuracy of 87% in predicting both the packet loss rate and the Mean Opinion Score (MOS). This approach not only enhances the understanding of call quality but also eliminates the need for listening to sensitive audio recordings made by third parties.

Key Contributions

Dataset Creation: As a part of the study, a specific dataset was created since no suitable dataset was available. The dataset used in the research has been shared on GitHub for future studies [2].
Machine Learning Model: A machine learning-based model was developed to predict the voice quality level and MOS score of recorded conversation sounds.

Future Directions

While the presented model has shown promising results, there are avenues for further exploration:

Dataset Expansion: Future studies can focus on expanding the dataset to improve the model's generalization and robustness.
Feature Extraction Methods: Exploring different approaches to feature extraction, such as alternative windowing methods for MFCCs, could enhance the model's performance.
Real-world Implementation: The proposed model could be tested and implemented in real-world VoIP systems to validate its effectiveness in practical scenarios.

Conclusion

Presenting at UBMK 2022 [3] was a rewarding experience, allowing me to share my research and engage with fellow professionals in the field. The positive reception and discussions surrounding the paper have inspired me to continue exploring innovative solutions for enhancing VoIP communication quality.

I look forward to future conferences and collaborative efforts to further advance the field of VoIP technology and quality prediction.