Search papers, labs, and topics across Lattice.
9
13
7
16
Factorized neural codec representations can achieve low-latency speech generation without sacrificing reconstruction quality.
Linguistic bias in spoofing detection can lead to a staggering 36.2% increase in error rates when models encounter out-of-domain data.
Kiwano revolutionizes speaker verification research by offering a comprehensive, user-friendly toolkit that streamlines experimentation and evaluation.
Naive data scaling can actually harm performance in voice spoofing detection, while just 8 hours of fine-tuning can dramatically boost robustness.
Speaker-invariant spoofing detection can be achieved without speaker labels, leading to a dramatic 25.7% improvement in detection accuracy across diverse datasets.
Current ASR metrics, even those leveraging embeddings, fail to align with human perception of transcription quality, as revealed by a new human-annotated dataset.
WER hides the real story: new metrics reveal how language model rescoring in ASR impacts grammatical correctness and semantic accuracy.
LLMs can judge speech recognition quality with near-human accuracy, blowing away traditional metrics like Word Error Rate.
Multi-corpus training can actually *hurt* spoofing detection, unless you strip out dataset-specific biases with this clever domain-invariant feature extraction trick.
