Developed and trained state-of-the-art models for Persian Automatic Speech Recognition, achieving a record-low 4.59% Word Error Rate (WER)—a 78% relative improvement over initial baselines. This work, which produced the best Persian ASR models to date, involved creating a massive 10,000-hour proprietary dataset and building scalable data processing pipelines.
Try the Demo
Published:
This project studies gibberish speech detection using phonetic and acoustic representations to distinguish clean speech from gibberish.
Published:
A multimodal speech emotion recognition study on IEMOCAP that combines frozen Wav2Vec2 and BERT representations with attention pooling and cross-modal attention.
Published:
An evaluation of whether explicitly adding emotion labels improves zero-shot depression detection on DAIC-WOZ transcripts.
Published in arXiv, 2025
Abstract: Automatic Speech Recognition (ASR) is an integral component of modern technology, powering applications such as voice-activated assistants, transcription services, and accessibility tools. Yet ASR systems continue to struggle with the inherent variability of human speech, such as accents, dialects, and speaking styles, as well as environmental interference, including background noise. Moreover, domain-specific conversations often employ specialized terminology, which can exacerbate transcription errors. These shortcomings not only degrade raw ASR accuracy but also propagate mistakes through subsequent natural language processing pipelines. Because redesigning an ASR model is costly and time-consuming, non-intrusive refinement techniques that leave the model’s architecture intact have become increasingly popular. In this survey, we review current non-intrusive refinement approaches and group them into five classes: fusion, re-scoring, correction, distillation, and training adjustment. For each class, we outline the main methods, advantages, drawbacks, and ideal application scenarios. Beyond method classification, this work surveys adaptation techniques aimed at refining ASR in domain-specific contexts, reviews commonly used evaluation datasets along with their construction processes, and proposes a standardized set of metrics to facilitate fair comparisons. Finally, we identify open research gaps and suggest promising directions for future work. By providing this structured overview, we aim to equip researchers and practitioners with a clear foundation for developing more robust, accurate ASR refinement pipelines.
Recommended citation: Peyghan, M. R., Soleimani Roudi, S., Zouashkiani, S., Amini, S., Rajabi, F., & Ghaemmaghami, S. (2025). "Non-Intrusive Automatic Speech Recognition Refinement: A Survey." arXiv.
Download Paper
Published in arXiv, 2026
Abstract: The advancement of zero-shot text-to-speech (TTS) synthesis is currently hindered for low-resource languages by the scarcity of large-scale, high-fidelity speech datasets. Traditional alignment-based methods require rare verbatim transcripts, while standard in-the-wild pipelines often rely on single-model automatic speech recognition (ASR) and silence-based segmentation, leading to transcription errors and truncated prosody. To address these challenges for the Persian language, this paper introduces PersianVox, a fully automated pipeline designed to generate high-quality speech corpora from unlabelled web data. Our approach integrates a novel prosody-aware segmentation strategy that utilizes acoustic turn-detection to preserve linguistic completeness and optimize utterance duration for long-context modeling. Furthermore, we employ a dual-ASR agreement mechanism, leveraging two distinct model architectures to filter unreliable transcriptions without ground truth. This pipeline yields a 1600-hour multi-speaker dataset, the largest open-source speech resource available for Persian to date. Additionally, we provide the first comparative benchmark of Speech Quality Assessment (SQA) methods for Persian, releasing a human-annotated subset to facilitate future research.
Recommended citation: Zouashkiani, S., Khalesi, S., Soleimani Roudi, S., Amini, S., & Ghaemmaghami, S. (2025). "PersianVox: A Prosody-Aware Dual-ASR Approach for Speech Corpus Generation from In-the-Wild Data." arXiv.
Download Paper
Hugging Face