acoustics

Iceland-based startup Treble has raised $18 million to advance its voice simulation platform for AI model training, hardware testing, and synthetic data generation, with customers including Amazon and Logitech.

Room reverberation and low-frequency noise from the environment hurt speech-to-text accuracy far more than the choice of model size; front-end audio preprocessing like adaptive spectral subtraction can recover masked phonemes and reduce word error rate more effectively than upgrading the model backend.

Researchers at UNSW Sydney have developed a new method to make espresso-strength coffee using ultrasonic sound waves and room-temperature water, reducing energy consumption by up to 75%. Blind taste tests showed the ultrasonic espresso is indistinguishable from traditionally brewed espresso.

Swanbench-Speech is a comprehensive benchmark for evaluating long-form speech generation across diverse scenarios, using multi-dimensional metrics covering acoustics, semantics, and expressiveness, revealing limitations of current models.