ADAPTIVE SPECTRAL NOISE REDUCTION TO IMPROVE SPEECH RECOGNITION

Oleh Osadchuk, Igor Olenych

Abstract


Background. Speech recognition technology is an important component of intelligent systems in various fields, in particular for voice control of smart home devices, personal identification in security systems, automation of various technological processes, etc. Increasing requirements for accuracy, speed, and autonomy of systems require improvement of speech recognition methods, especially when affected by background noise. Therefore, the goal of this work was to develop an adaptive noise reduction method to improve the accuracy of audio information recognition.

Materials and methods. The study of the impact of noise environments on the accuracy of automatic speech recognition covers the stages of spectral analysis of audio signals using short-term Fourier transformation, adaptive noise reduction using soft masks and Wiener filter elements, and signal reconstruction based on inverse Fourier transform. The study uses Whisper speech recognition models and Common Voice and LibriSpeech datasets, supplemented with synthesized noisy recordings.

Results and Discussion. It has been established that various types of background noise (specifically low-frequency stationary, percussive, impulse, tonal, and broadband) and signal-to-noise ratios affect the speech recognition accuracy of Whisper models differently. A method to improve speech recognition accuracy based on adaptive noise reduction is proposed. This method provides flexible suppression of unwanted frequencies without significant speech distortion and allows for the compensation of even complex background noises. A reduction in the average consonant confusion level by 50–60% for the Whisper Tiny model and by 15–30% for the Whisper-1 model in real-time has been demonstrated.

Conclusion. As a result of analyzing the speech recognition efficiency of Whisper models based on phonetic distortion metrics, transcription accuracy, consonant confusion, and artifact perception across a wide range of signal-to-noise ratios, it has been established that the application of adaptive noise reduction provides a significant improvement in accuracy. Despite the higher accuracy of the Whisper-1 model, the Whisper Tiny model with a soft mask demonstrates sufficient efficiency for intelligent systems based on real-time IoT platforms.

Keywords: Whisper models, background noise, noise reduction, Wiener filter, spectral masking, IoT.


Full Text:

PDF

References


[1] Smart Secure Homes: A Survey of Smart Home Technologies that Sense, Assess, and Respond to Security Threats / J. Dahmen, D. J. Cook, X. Wang, W. Honglei. Journal of Reliable Intelligent Environments. 2017. Vol. 3, no. 2. P. 83–98. https://doi.org/10.1007/s40860-017-0035-0

[2] Knowledge-based Cyber Physical Security at Smart Home: A Review / A. Alsufyani, O. Rana, C. Perera. ACM Computing Surveys. 2024. Vol. 57, iss. 3. Art. 53. P. 1–36. https://doi.org/10.1145/3698768

[3] Noise Reduction in Industry Based on Virtual Instrumentation / R. Martinek, R. Jaros, J. Baros [et al.]. Computers, Materials and Continua. 2021. Vol. 69, iss. 1. P. 1073–1096. https://doi.org/10.32604/cmc.2021.017568

[4] Real-Time Speech Recognition for IoT Purpose using a Delta Recurrent Neural Network Accelerator / C. Gao, S. Braun, I. Kiselev, J. Anumula. 2019 IEEE International Symposium on Circuits and Systems (ISCAS) : Conference Paper. 2019. https://doi.org/10.1109/ISCAS.2019.8702290

[5] Application of speech recognition technology in IoT smart home / P. Wang, X. Lu, H. Sun, W. Lv. 2019 IEEE 3rd Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC) : Conference Paper. 2019. https://doi.org/10.1109/IMCEC46724.2019.8984175

[6] Towards Energy-Efficient and Low-Latency Voice-Controlled Smart Homes: A Proposal for Offline Speech Recognition and IoT Integration / P. Huang, I. Ullah, X. Wei, T. A. Ahanger. arXiv. 2025. https://doi.org/10.48550/arXiv.2506.07494

[7] Whisper-AT: Noise-Robust Automatic Speech Recognizers are Also Strong General Audio Event Taggers / Y. Gong, S. Khurana, L. Karlinsky, J. Glass. Proceedings of the 24th Annual Conference of the International Speech Communication Association (INTERSPEECH 2023). 2023. P. 2798–2802. https://doi.org/10.21437/Interspeech.2023-2193

[8] Improving Speech Recognition Performance in Noisy Environments by Enhancing Lip Reading Accuracy / D. Li, Y. Gao, C. Zhu [et al.]. Sensors. 2023. Vol. 23, iss. 4. Art. 2053. https://doi.org/10.3390/s23042053

[9] AI-Driven Real-Time Distress Detection Through Speech Recognition for Emergency Response Systems / M. Halilaj, E. Myrto, A. F. Dragoni. CEUR Workshop Proceedings (CEUR-WS.org). 2025. Vol. 4044, paper 03. https://ceur-ws.org/Vol-4044/paper03.pdf

[10] IoT based speech recognition system / K. K. Sethy, L. M. Varalakshmi, R. E. [et al.]. AIP Conference Proceedings. 2022. Vol. 2393, iss. 1. Art. 020096. https://doi.org/10.1063/5.0074140

[11] Unsupervised Spectral Subtraction for Noise-Robust ASR / G. Lathoud, M. Magimai-Doss, B. Mesot, H. Bourlard. IDIAP Research Institute. 2006. https://publications.idiap.ch/downloads/papers/2005/lathoud05e.pdf

[12] Musical Noise Reduction By Processing Spectrogram of Spectral Subtraction Output / H. R. Abutalebi, B. Dashtbozorg. Proceedings of the 15th International Congress on Sound and Vibration (ICSV15). 2008. P. 2979–2986. https://www.researchgate.net/publication/228916452_Signal_Processing_Musical_Noise_Reduction_By_Processing_Spectrogram_of_Spectral_Subtraction_Output

[13] Vaseghi S. V. Advanced Digital Signal Processing and Noise Reduction. 2nd ed. John Wiley & Sons Ltd, 2000. 488 p. ISBN 0-471-62692-9. https://scispace.com/pdf/advanced-digital-signal-processing-and-noise-reduction-375hxslkau.pdf

[14] Wiener Filter and Deep Neural Networks: A Well-Balanced Pair for Speech Enhancement / D. Ribas, A. Miguel, A. Ortega, E. Lleida. Applied Sciences. 2022. Vol. 12, iss. 18. Art. 9000. https://doi.org/10.3390/app12189000

[15] Reddy A., Raj B. Soft Mask Methods for Single-Channel Speaker Separation. IEEE Transactions on Audio, Speech, and Language Processing. 2007. Vol. 15, iss. 6. P. 1766–1776. https://doi.org/10.1109/TASL.2007.901310

[16] Hout J. van, Alwan A. A novel approach to soft-mask estimation and Log-Spectral enhancement for robust speech recognition. 2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 2012. P. 4105–4108. https://doi.org/10.1109/ICASSP.2012.6288821

[17] ] Radfar M. M., Dansereau R. M. Single-Channel Speech Separation Using Soft Mask Filtering. IEEE Transactions on Audio, Speech, and Language Processing. 2007. Vol. 15, iss. 8. P. 2299–2310. https://doi.org/10.1109/TASL.2007.904233




DOI: http://dx.doi.org/10.30970/eli.34.11

Refbacks

  • There are currently no refbacks.