This paper analyzes speech-text interleaved language models and reveals that they go through an implicit transcription phase in which spoken words become decodable as text in intermediate layers.

PolySpeech-100 introduces a massive, multi-lingual benchmark covering 110 linguistic variants to rigorously test Speech-LLMs, demonstrating that open-source models struggle with low-resource languages…

This paper explores the effect of conversational timing properties on automatic speech recognition (ASR) systems by controlling and optimizing pause and overlap timing distributions.

The paper proposes RTFree-F5, a method to make flow-matching TTS models like F5-TTS independent of reference transcripts, improving performance and naturalness for dysarthric speakers.

This paper investigates the relationship between language model perplexity and word error rate in modern automatic speech recognition systems, studying the impact of external language models, encoder…

A lightweight framework for automated pronunciation assessment using native speech resources and unsupervised or lightly calibrated methods.

The paper introduces GigaSpeechBench, a comprehensive multilingual and multidimensional ASR & AST benchmark with 680 hours of human-annotated speech, featuring 12 low-resource languages, 6 Chinese dia…

This paper proposes four guidelines and two novel data ordering methods (STR and SAW) to systematically optimize data organization, significantly enhancing the stability and performance of LLM trainin…

The paper introduces Script-Normalized WER (SN-WER), a novel evaluation metric that transliterates ASR transcripts into a canonical script to accurately measure speech recognition performance across d…

This paper uses continual learning with explicit disfluency tokens to improve Automatic Speech Recognition (ASR) systems on disfluent speech, addressing the information loss and hallucinations caused…

The paper proposes Agentic ASR, a closed-loop framework that treats ASR as a multi-turn refinement task, significantly improving semantic accuracy over traditional token-level metrics.

The paper introduces OpenSTBench, a unified, multidimensional evaluation framework designed to comprehensively compare heterogeneous speech translation systems by jointly assessing translation, speech…

The paper introduces BEA-Dialogue+, an expanded 200-hour corpus for Hungarian conversational ASR, demonstrating that while larger data is challenging, specialized fine-tuning techniques significantly…

CORTIS is a text-only adaptation framework that fine-tunes spoken language models for task-oriented voice agents using text-form task supervision.