The paper introduces a compact BART-based sequence‑to‑sequence model for low‑latency spell correction of Japanese music search queries, addressing challenges posed by four writing scripts. It employs a script‑aware synthetic misspelling pipeline that blends keyboard‑layout models, phonetic confusion priors, consonant alternations, and kana case errors, and normalizes mixed‑script titles to a single canonical script to reduce hallucinations. Experiments show the model achieves 41.09% exact‑match accuracy and 11.62% CER, outperforming edit‑distance baselines while keeping inference latency under 4 ms on a single GPU.
While large language model (LLM)-based text-to-speech (TTS) systems have achieved high-quality speech synthesis, most existing systems focus on English and Chinese. Japanese, however, remains under-explored, and its unique linguistic challenges, such as widespread context-dependent kanji polyphony, have yet to be adequately tackled.
The paper introduces GUIDE, a generative unsupervised framework for Chinese query correction that uses shared-ID encoding for phonetically or visually confusable characters and an encoder–decoder architecture to reconstruct queries within plausible confusion neighborhoods. It incorporates a time‑decayed, query‑frequency‑weighted objective to adapt to rapidly changing query vocabularies. Experiments on QSpell 250K and the large‑scale KwaiSearch dataset demonstrate that GUIDE consistently outperforms strong baselines, with online A/B testing confirming improvements in correction quality and downstream engagement.
The paper presents LMSpell, an empirical study evaluating how different pre‑trained language models (PLMs) perform at spell correction across multiple languages, including low‑resource languages (LRLs). It demonstrates that even relatively small PLMs such as Gemma 3 (270 M parameters) and mBART50, when fine‑tuned on only 5,000 sentences, can surpass rule‑based spell correctors. A case study on Sinhala illustrates the challenges of spell correction for LRLs.
The paper introduces a new benchmark of 13,044 spontaneous Nepali‑English code‑switched podcast utterances, totaling 15.45 h, and evaluates a two‑stage synthetic‑to‑real adaptation framework for ASR. Stage 1 uses supervised fine‑tuning on 0.95 h of synthetic speech, reducing Mixed Error Rate from 23.40 % to 21.52 % and switch‑point error rate from 35.89 % to 28.77 %. Stage 2 applies Group Relative Policy Optimization with script and transcription rewards, yielding modest but statistically significant further improvements, particularly in English script confusion and switch‑point accuracy. The study also shows that an auxiliary script reward term offers no benefit beyond MER‑only optimization in this setting, and that the synthetic SFT provides the main cross‑domain transfer to spontaneous human speech.
arXiv:2608.22872v1 Announce Type: new Abstract: Speech-based applications pass spoken queries through automatic speech recognition (ASR) before any retrieval module, so ASR errors enter the pipeline...
arXiv:2608.22872v2 Announce Type: new Abstract: Speech-based applications pass spoken queries through automatic speech recognition (ASR) before any retrieval module, so ASR errors enter the pipeline...
Chinese news text contains dense written forms such as scores, hyphenated model names, ranges, unit symbols, percentages, English abbreviations, and mixed Chinese-Latin-digit names. These forms are frequent in real listening workflows, and a text-to-speech (TTS) system can preserve the written string while changing the spoken meaning.
The paper introduces a unified phoneme‑based TTS‑to‑ASR augmentation pipeline that uses a multilingual TTS model with language‑ID conditioning and incorporates grapheme‑to‑phoneme conversion, reference‑speech filtering, and candidate‑text selection. It proposes phoneme‑frequency‑guided selection (PFGS) to rank sentences based on phoneme frequencies from real ASR labels, and demonstrates that random augmentation and PFGS both improve ASR performance across Arabic, French, Italian, and Portuguese test sets, with PFGS yielding up to a 19.3% relative WER reduction. The study also shows that filtering reference speech can further lower WER by up to 0.59 points on certain datasets.
arXiv:2606. 20518v1 Announce Type: new Abstract: Flow-matching text-to-speech systems achieve remarkable zero-shot quality but remain static after deployment: pronunciation errors on out-of-vocabulary proper nouns persist unless the model is retrained.
The paper introduces a training‑free speech‑and‑text‑to‑pronunciation (ST2P) pipeline that combines lexical candidates from G2P tools with acoustic rescoring using frozen pretrained S2P models. By performing a left‑to‑right greedy search over whole‑sequence negative log‑likelihoods, the method achieves a dramatic reduction in character error rate on Japanese corpora, outperforming both baseline G2P/S2P approaches and commercial multimodal LLMs. The approach is also significantly faster—3–3.5× faster than beam search and twice as fast as direct decoding—while maintaining high accuracy across multiple languages.
The paper examines how different data‑augmentation techniques affect low‑resource automatic speech recognition beyond overall error rates. It compares speed perturbation, SpecAugment, and two text‑based audio synthesis methods across four low‑resource languages and English, analyzing phoneme‑level errors. The study finds that SpecAugment improves aggregate performance but increases tone and palatalization errors, while text‑based methods introduce consonant insertions; when combined, each method mitigates the other's weaknesses, yielding superadditive gains that diminish as data size grows to 100 hours.
arXiv:2606. 17826v1 Announce Type: cross Abstract: Automatic speech recognition (ASR) in non-English clinical settings is challenged by multiscript variability, where the same term may appear in multiple valid orthographic forms.