Search papers, labs, and topics across Lattice.
7
5
4
Real-world speech recordings with overlapping noise events are now timestamped and categorized, filling a crucial gap in audio datasets.
Out-of-domain generalization in speech encoders hinges on their proximity to unseen TTS embeddings rather than their distance from natural speech.
Joint language-district supervision not only boosts district discrimination but also preserves language classification integrity, revealing a nuanced structure in speech embeddings.
Geographic distance significantly predicts ASR performance, revealing that models struggle with regional variations in Indian languages.
FastConformer achieves over 90% accuracy on out-of-domain language identification, outperforming Whisper without task-specific adaptation.
ASR systems exhibit surprising language-specific sensitivities, revealing that speaker behavior and signal processing choices can drastically affect performance across Indic languages.
VAANI's open-sourced dataset offers unprecedented coverage of India's linguistic landscape, finally giving researchers the data needed to build truly inclusive speech models.
