Topic
audio-processing
Improving Code-Switching ASR with Code-Mixing Guided Synthetic Speech
A new code-mixing guided preference-learning framework improves code-switching automatic speech recognition (ASR) by steering synthetic speech generation toward better language-boundary consistency. Fine-tuning Whisper Large with this approach reduced Mixed Error Rate (MER) from 12.1% to 8.9% on the DevMAN set and from 17.8% to 14.2% on the DevSGE set of the SEAME Mandarin-English conversational corpus.
Improving End-to-End Speech Recognition for Dysarthric Speech through In-Domain Data Augmentation
A study by Sapkota et al. explores data augmentation techniques for dysarthric automatic speech recognition (ASR) by fine-tuning the end-to-end Wav2Vec2 model. Four methods—Speaking-Rate Modification, Pitch Modification, Formant Modification, and vocal tract Length Perturbation—were tested across severity levels, achieving relative WER reductions of 30.02%, 16.64%, and 15.47% for low, medium, and high severity respectively.
New Temporal Pyramid Model Enhances Spoofed Speech Detection for Voice Security Systems
Researchers introduced a Temporal Pyramid Adapter for spoofed speech detection that uses parallel temporal convolutions with varying receptive fields to capture multi-scale cues. The model achieved a 99.24% AUC and 3.87% EER on the PartialSpoof dataset, significantly outperforming existing methods like LCNN-BLSTM (9.87% EER) and TRACE (8.08% EER). The work highlights the potential for improving voice authentication security but notes performance degradation under domain and language shifts.
Neural Audio Codecs' Low Frame Rate Degradation Linked to Training Configuration
A new study by Gichamba and Busogi investigates the mechanisms behind low frame rate degradation in neural audio codecs. The researchers found that a quality cliff at 6.25 Hz is caused by suboptimal training configuration, not by phonemic collisions or codebook saturation. After correcting the training setup, the codecs perform smoothly down to 3.1 Hz and 1.6 Hz, suggesting that low frame rate efficiency gains are more accessible than previously assumed.
Prototype Adaptation and Pseudo Class-Variable Training Boost Few-Shot Audio Classification
Researchers propose a method for few-shot class-variable incremental audio classification, handling both increases and decreases in the number of classes. The approach uses a prototype adaptation network and pseudo class-variable training. Experiments on three public datasets show improved average accuracy over previous methods.
NVMOS: Novel AI Model Predicts Perceptual Quality of Non-Verbal Vocalizations in Speech
A new paper on arXiv introduces NVMOS, the first model purpose-built to assess the perceptual quality of non-verbal vocalizations (NVs) such as laughter, sighs, and coughs in speech. The model was trained on a newly constructed NV-MOS dataset with expert ratings and achieves expert-level agreement with human Mean Opinion Scores. Tests on multimodal LLMs like Gemini showed clear inconsistencies, highlighting the need for specialized NV quality assessment.