Tag
This paper proposes using self-supervised speech representations from HuBERT to track the convergence of children's speech to adult patterns in deaf and hard-of-hearing infants, demonstrating a scalable, language-neutral assessment method for language development.
This paper introduces PINT, a method for invariant speech tokenization that fine-tunes an SSL encoder using alignment losses across parallel utterances to distill linguistic content, achieving significant reductions in speaker probe accuracy and LM perplexity.
Proposes speaker-disentangled syllabic tokenization using chunk-wise regression to learn linguistic content tokens from raw speech, achieving state-of-the-art in syllable boundary detection and clustering, and improving speech language model performance.
This paper proposes a speaker-disentangled syllabic tokenizer that uses regression of perturbed student representations toward clean teacher targets, achieving state-of-the-art syllable boundary detection and a 7% relative improvement in speech language model understanding over SpiRit-LM.
This paper presents a systematic empirical study of fine-tuning pretrained Transformer models (Wav2Vec2.0, HuBERT, XLS-R) for Quranic Automatic Speech Recognition (ASR), achieving a WER of 0.08 on the EveryAyah subset and reducing training time from 140 to 40 hours, with Wav2Vec2-XLSR-53 providing the best representation.
A C++ implementation of distilHuBERT with no runtime dependencies, compiled-in weights, dynamic sizing, and on-par performance with ONNX Runtime, designed for easy integration into CMake projects.