A new framework shows concept subspaces are not unique, estimator choice affects containment and disentanglement, LEACE works well but generalizes poorly, and HuBERT encodes phone info as contained and disentangled from speaker info while speaker info resists compact containment.
Scaling speech technology to 1,000+ languages
11 Pith papers cite this work, alongside 116 external citations. Polarity classification is still indexing.
11
Pith papers citing it
116
external citations · external index
representative citing papers
Tadabur is a large-scale Quran audio dataset with over 1400 hours from 600+ reciters to support speech research and benchmarks.
A training-free method quantifies dysarthria severity via d-prime scores on phonological contrasts in HuBERT embeddings, correlating with clinical ratings across 5 languages and multiple conditions.
Multilingual ASR models show 39.7-297% zero-shot WER on Pashto public data, Whisper models output correct script in under 0.8% of cases, and fine-tuned models degrade to 32.5-59% WER on out-of-domain sets.
BlasBench supplies an Irish-aware normalizer and scoring harness that enables reproducible ASR comparisons and exposes a 33-43 point generalization gap for fine-tuned models versus 7-10 points for massively multilingual ones.
RuASD is a comprehensive Russian speech anti-spoofing dataset featuring 37 synthesis systems and a robustness evaluation pipeline for real-world channel distortions.
Language coherence arises from slow contextual integration in default-mode cortex and rapid event-driven reconfiguration in auditory and language areas, captured by LLM-derived signals in single-subject fMRI.
MLAAD provides a large-scale multi-language synthetic audio dataset for training and evaluating audio anti-spoofing models, showing better training performance than InTheWild and FakeOrReal and alternating superiority with ASVspoof 2019 across eight test sets.
Frozen FastConformer with hierarchical softmax achieves over 90% macro accuracy on out-of-domain Indic LID benchmarks for 42 languages and outperforms Whisper and other objectives in cross-corpus settings.
A shared LoRA adapter on VoxCPM2 trained on 26 hours of Khmer and Korean data improves Khmer MOS from 3.85 to 4.23 using under 3% of parameters, with no gain for Korean.
Fine-tuned Wav2Vec2 and HuBERT models recognize click consonants more accurately than non-clicks in G|ui and West !Xoon data.
citing papers explorer
-
A framework for analyzing concept representations in neural models
A new framework shows concept subspaces are not unique, estimator choice affects containment and disentanglement, LEACE works well but generalizes poorly, and HuBERT encodes phone info as contained and disentangled from speaker info while speaker info resists compact containment.
-
Tadabur: A Large-Scale Quran Audio Dataset
Tadabur is a large-scale Quran audio dataset with over 1400 hours from 600+ reciters to support speech research and benchmarks.
-
Training-Free Cross-Lingual Dysarthria Severity Assessment via Phonological Subspace Analysis in Self-Supervised Speech Representations
A training-free method quantifies dysarthria severity via d-prime scores on phonological contrasts in HuBERT embeddings, correlating with clinical ratings across 5 languages and multiple conditions.
-
Benchmarking Multilingual Speech Models on Pashto: Zero-Shot ASR, Script Failure, and Cross-Domain Evaluation
Multilingual ASR models show 39.7-297% zero-shot WER on Pashto public data, Whisper models output correct script in under 0.8% of cases, and fine-tuned models degrade to 32.5-59% WER on out-of-domain sets.
-
BlasBench: An Open Benchmark for Irish Speech Recognition
BlasBench supplies an Irish-aware normalizer and scoring harness that enables reproducible ASR comparisons and exposes a 33-43 point generalization gap for fine-tuned models versus 7-10 points for massively multilingual ones.
-
Evaluating Generalization and Robustness in Russian Anti-Spoofing: The RuASD Initiative
RuASD is a comprehensive Russian speech anti-spoofing dataset featuring 37 synthesis systems and a robustness evaluation pipeline for real-world channel distortions.
-
Coherence in the brain unfolds across separable temporal regimes
Language coherence arises from slow contextual integration in default-mode cortex and rapid event-driven reconfiguration in auditory and language areas, captured by LLM-derived signals in single-subject fMRI.
-
MLAAD: The Multi-Language Audio Anti-Spoofing Dataset
MLAAD provides a large-scale multi-language synthetic audio dataset for training and evaluating audio anti-spoofing models, showing better training performance than InTheWild and FakeOrReal and alternating superiority with ASVspoof 2019 across eight test sets.
-
A Comparative Study of Pre-trained Speech Encoders and Training Objectives for Large-Scale Indic Spoken Language Identification
Frozen FastConformer with hierarchical softmax achieves over 90% macro accuracy on out-of-domain Indic LID benchmarks for 42 languages and outperforms Whisper and other objectives in cross-corpus settings.
-
Closing the Quality Gap in Low-Resource Text-to-Speech: LoRA Fine-Tuning of VoxCPM2 for Khmer and Korean
A shared LoRA adapter on VoxCPM2 trained on 26 hours of Khmer and Korean data improves Khmer MOS from 3.85 to 4.23 using under 3% of parameters, with no gain for Korean.
-
Pretrained self-supervised speech models can recognize unseen consonants
Fine-tuned Wav2Vec2 and HuBERT models recognize click consonants more accurately than non-clicks in G|ui and West !Xoon data.