Search papers, labs, and topics across Lattice.
Automatic music transcription models struggle with pop music, as evidenced by a mere 38% Onset F1 score on the new MulTTiPop dataset.
Timestamp drift in ASR can be corrected with minimal parameter updates, achieving near-perfect alignment without sacrificing model performance.
Decomposer achieves superior MIDI reconstruction fidelity and code readability compared to existing models, transforming how we approach symbolic music decompilation.
Achieving 80% throughput with Classifier-Free Guidance challenges the assumption that CFG drastically reduces efficiency in multimodal audio processing.
Better attribution in generative music could significantly boost creator welfare and reshape platform compensation strategies.
Personalized fine-tuning of ASR models can reduce word error rates for dysarthric speech to as low as 9.7%, transforming communication for affected individuals.
Current multimodal conversational models miss critical emotional cues, revealing a significant gap in their ability to engage in nuanced human-like dialogue.
LALMs can achieve superior emotional comparison accuracy with only 5% of the training data typically required by conventional methods.
Speech synthesis can now adaptively enhance clarity and vocal effort, mimicking human responses in noisy environments.
Bagpiper-TTS can seamlessly transform natural language requests into high-quality speech across diverse applications, outperforming traditional TTS systems.
Expert iteration emerges as the key driver of quality in text-to-music generation, overshadowing the contributions of preference tuning.
Weak audio supervision allows ReNikud to achieve superior grapheme-to-phoneme conversion for Hebrew, outperforming traditional methods that struggle with data scarcity and pronunciation accuracy.
Idempotency in training voice attribute editing models can drastically reduce the impact of noisy labels, leading to more reliable and consistent edits.
S-JEPA sets a new standard in speech representation learning by achieving top performance with fewer parameters and without the cumbersome offline re-clustering process.
Dixtral achieves up to 29% absolute improvement in speaker-attributed transcription accuracy by leveraging diarization masks without risking catastrophic forgetting.
The quality cliff at low frame rates is driven by training configuration issues, not inherent limitations of neural audio codecs.
TuneJury achieves superior music preference alignment with a single frozen reward model that adapts efficiently to new audio generators.
Reducing inter-utterance silence from 9.6 seconds to 0.3 seconds transforms the quality of real-time game commentary, making it feel more natural and engaging.
Anticipating dialogue endpoints up to 2.56 seconds ahead can slash latency by over half while enhancing computational efficiency in real-time speech interactions.
A groundbreaking dataset of 313 hours of real-world code-switched speech reveals rich patterns and frequencies previously overlooked in multilingual research.
Incremental speech quality assessment can be dramatically improved by modeling it as a multi-resolution task, achieving a 48% reduction in error on partial audio inputs.
Only half of speech translation interactions are rated as usable, revealing critical usability gaps that standard evaluations overlook.
Exploiting noise inseparability allows for a breakthrough in weakly-supervised speech denoising, enhancing domain adaptability and performance without relying on clean targets.
Training speech separation models on real-world noisy data doesn't have to mean accepting noisy outputs: this method cuts residual noise in half.
Open-ear smart glasses can now achieve >11dB noise reduction with a real-time active noise cancellation system, despite lacking a sealed ear canal.
Forget hand-tuning: this recipe for universal phone recognition leverages large-scale multilingual data and SSL to achieve SOTA performance across 100+ languages.
Finally, digital humans can have realistic, socially aware conversations: DyaDiT generates dyadic gestures that users strongly prefer over existing methods.
