arXiv subjects

Mohammad Sadegh Safari

Publications and source records attributed to Mohammad Sadegh Safari.

4 recordsLinked to original sources

arXiv2025

PSRB: A Comprehensive Benchmark for Evaluating Persian ASR Systems

Although Automatic Speech Recognition (ASR) systems have become an integral part of modern technology, their evaluation remains challenging, particularly for low-resource languages such as Persian. This paper introduces Persian Speech Recognition Benchmark(PSRB), a comprehensive benchmark designed to address this gap by incorporating diverse linguistic and acoustic conditions. We evaluate ten ASR systems, including state-of-the-art commercial and open-source models, to examine performance variations and inherent biases. Additionally, we conduct an in-depth analysis of Persian ASR transcriptions, identifying key error types and proposing a novel metric that weights substitution errors. This metric enhances evaluation robustness by reducing the impact of minor and partial errors, thereby improving the precision of performance assessment. Our findings indicate that while ASR models generally perform well on standard Persian, they struggle with regional accents, children's speech, and specific linguistic challenges. These results highlight the necessity of fine-tuning and incorporating diverse, representative training datasets to mitigate biases and enhance overall ASR performance. PSRB provides a valuable resource for advancing ASR research in Persian and serves as a framework for developing benchmarks in other low-resource languages. A subset of the PSRB dataset is publicly available at https://huggingface.co/datasets/PartAI/PSRB.

eess.AS↗

arXiv2025

End-to-End Target Speaker Speech Recognition Using Context-Aware Attention Mechanisms for Challenging Enrollment Scenario

This paper presents a novel streaming end-to-end target-speaker speech recognition that addresses two critical limitations in systems: the handling of noisy enrollment utterances and specific enrollment phrase requirements. This paper proposes a robust Target-Speaker Recurrent Neural Network Transducer (TS-RNNT) with dual attention mechanisms for contextual biasing and overlapping enrollment processing. The model incorporates a text decoder and attention mechanism specifically designed to extract relevant speaker characteristics from noisy, overlapping enrollment audio. Experimental results on a synthesized dataset demonstrate the model's resilience, maintaining a Word Error Rate (WER) of 16.44% even with overlapping enrollment at 5dB Signal-to-Interference Ratio (SIR), compared to conventional approaches that degrade to WERs above 75% under similar conditions. This significant performance improvement, coupled with the model's semi-text-dependent enrollment capabilities, represents a substantial advancement toward more practical and versatile voice-controlled devices.

eess.AS↗

arXiv2021

Progressive Transmission using Recurrent Neural Networks

In this paper, we investigate a new machine learning-based transmission strategy called progressive transmission or ProgTr. In ProgTr, there are b variables that should be transmitted using at most T channel uses. The transmitter aims to send the data to the receiver as fast as possible and with as few channel uses as possible (as channel conditions permit) while the receiver refines its estimate after each channel use. We use recurrent neural networks as the building block of both the transmitter and receiver where the SNR is provided as an input that represents the channel conditions. To show how ProgTr works, the proposed scheme was simulated in different scenarios including single/multi-user settings, different channel conditions, and for both discrete and continuous input data. The results show that ProgTr can achieve better performance compared to conventional modulation methods. In addition to performance metrics such as BER, bit-wise mutual information is used to provide some interpretation to how the transmitter and receiver operate in ProgTr.

eess.SP↗

arXiv2019

Deep UL2DL: Channel Knowledge Transfer from Uplink to Downlink

Knowledge of the channel state information (CSI) at the transmitter side is one of the primary sources of information that can be used for the efficient allocation of wireless resources. Obtaining downlink (DL) CSI in Frequency Division Duplexing (FDD) systems from uplink (UL) CSI is not as straightforward as in TDD systems. Therefore, users usually feed the DL-CSI back to the transmitter. To remove the need for feedback (and thus having less signaling overhead), we propose to use two recent deep neural network structures, i.e., convolutional neural networks and generative adversarial networks (GANs) to infer the DL-CSI by observing the UL-CSI. The core idea of our data-driven scheme is exploiting the fact that both DL and UL channels share the same propagation environment. As such, we extracted the environment information from the UL channel response to a latent domain and then transferred the derived environment information from the latent domain to predict the DL channel. To overcome incorrect latent domain and the problem of oversimplistic assumptions, in this work, we did not use any specific parametric model and instead used data-driven approaches to discover the underlying structure of data without any prior model assumptions. To overcome the challenge of capturing the UL-DL joint distribution, we used a mean square error-based variant of the GAN structure with improved convergence properties called boundary equilibrium GAN (BEGAN). For training and testing we used simulated data of Extended Vehicular-A (EVA) and Extended Typical Urban (ETU) models. Simulation results verified that our methods can accurately infer and predict the downlink CSI from the uplink CSI for different multipath environments in FDD communications.

eess.SP↗