Muhammadjon Musaev

To generate each token, the Whisper decoder re-reads the encoder’s 1500-position cross-attention key–value (KV) cache on the CPU. In the Whisper-medium model, this cache occupies 281 MiB; for short recordings, 74–82% of it consists of padding positions. This paper proposes a calibratable framework that reduces the number of KV positions read at each step without retraining the model or changing the cache kept in memory. The saving therefore comes from reading fewer bytes from DRAM at each step. Its core method—PadSink-Track—preserves critical “padding-sink” positions, shifts the audio window at each step based on signals from Whisper’s alignment heads, and updates selected KV entries in a ring buffer. A fallback mechanism re-executes the step using the full cache when confidence is low. The budget, sink fraction, and fallback threshold (τ) are calibrated for each model on a validation set based on a specified non-inferiority criterion (δ = 0.2 × WER). When the calibrated budget was halved, none of the selection rules relying solely on the first step met the criterion. In contrast, the calibrated framework satisfied this criterion across all ten configurations, covering four languages (Uzbek, English, Russian, Turkish) and two model sizes. On the held-out sets, every no-fallback decision was reproduced, and the single fallback decision (medium/English) became “Inconclusive”. In three configurations, where 19–35% of the steps were re-executed, the speed condition selected the full cache. On the Intel Core i7-11850H processor, decoder-step time without fallback decreased by 32–34% (synthetic step; 23–25% with real decoding), the volume of data read from DRAM dropped by 36–40%, and the throughput of the eight parallel decoders increased by 53% (synthetic step; 33% with real decoding, median latency −26%). Time savings during whole-utterance decoding amounted to 13–14% for models without fallback, whereas for the original medium model that required fallback, this figure was approximately 4%.

In post-training compression of Transformer models, the compression ratio or pruning rate is often predetermined and does not directly take into account the memory capabilities of the target hardware. This work introduces a cache-aware post-training compression framework for CPU-class hardware. The framework derives the minimum required compression ratio from the effective weight working set of the operator or layer and the effective level-3 (L3) cache budget. The cache budget is treated as a hardware resource shared with concurrent modules rather than as a latency heuristic. If functional redundancy is present, functionally similar channels are physically removed using compensated structured pruning; in cases where final 8-bit integer (INT8) alone cannot meet the hardware target, activation-aware low-rank factorization is used. The resulting weights are quantized to per-output-channel INT8 using GPTQ, and the final configuration is evaluated by a task-level accuracy gate. In the Whisper-medium Uzbek automatic speech recognition (ASR) encoder, removing 17.1% of the feed-forward network (FFN) channels reduced the size from 300 MiB to 267 MiB, with the word error rate (WER) difference from the GPTQ-only variant (per-output-channel INT8 without structured pruning) not statistically distinguishable (ΔWER = −0.0014, 95% confidence interval (CI) [−0.0115, +0.0096]). For the full model, the cascade achieved 4.14× compression at 705 MiB with WER = 0.1833, whereas extending compression to 5.34× increased WER to 0.6101. Hardware profiling showed 2.41× fewer memory stalls and a 1.91× reduction in encoder-plus-decoder-step execution time. End-to-end response time improved by 2.36×, reducing the real audio RTF from 2.57 to 1.09. Taken together, these results support hardware-derived targeting with accuracy-guided stopping as an alternative to selecting a fixed compression ratio in advance.

The relevance of creating information systems using artificial intelligence methods and tools is dictated by the following reasons: The volume of scientific and educational information is growing; Traditional information retrieval methods have exhausted themselves. Using only deterministic and iteration methods, rigid algorithms don't give the expected results. They require more time to process information and more memory. Significant progress in recent years in the development of artificial intelligence (AI) methods and systems gives hope that their use will significantly reduce the time needed to search for data for scientific research and educational activities. The aim of the research results presented in the article is to increase the efficiency for scientific and educational information retrieval based on the use of AI methods implemented in the integrated intelligent information system “SMART TUIT". The article presents the results of theoretical and applied research obtained by several departments of the Tashkent University of Information Technology (TUIT) in solving the following tasks: Voice recognition for subsequent processing; Pattern recognition in order to identify the users of information; Search and processing of scientific and educational resources in electronic libraries; Analysis of information needs of users depending on the level of competence and type of activity; Evaluation of scientific and educational information to identify the most important data sources; Geoinformation system to solve the problems of the location of the information source. Initially, each research area in the departments was aimed at solving a certain class of problems related to medicine, linguistics, electronic libraries, corporate networks, information security systems, etc. The TUIT creative group decided to combine efforts to apply the results obtained to solve the important problem of intellectualizing the search for sources of scientific and educational information among a large amount of data.

In this paper has been discussed about speech recognition using spectrogram images and deep convolution neural network(CNN) of Uzbek spoken digits. Spectrogram images from speech signal were generated and it were used for deep CNN training. Presented CNN model contains 3 convolution layers and 2 fully connected layers that discriminative features can be divided and estimated of spectrogram images by those layers. In current research period, dataset of Uzbek spoken digits were made and in based on presented CNN model they were trained. Testing results shows that, proposed approach for Uzbek spoken digits classified 100% accuracy.