Abstract: Handcrafted audio descriptors and learned deep representations each bring distinct strengths and inherent limitations to speech emotion recognition (SER). Traditional handcrafted features ...

Abstract: Despite recent advancements in speech processing, zero-resource speech translation (ST) and automatic speech recognition (ASR) remain challenging problems. In this work, we propose to ...

Note: OpenVINO is currently incompatible with Kokoro models due to dynamic rank tensor requirements. The provider will automatically fall back to CPU if OpenVINO fails.