Voicebox Releases Talisman Embedded Speech Recognition
Voicebox Technologies, a provider of natural language voice applications, has released Talisman, the latest iteration of its Embedded Automatic Speech Recognition (ASR) product for automotive and embedded IoT applications.
Talisman ASR is designed for small and mid-range embedded platforms, like those found in cars, smartphones, and smart appliances. It supports large recognition grammars that can handle thousands of phrases.
Historically, Voicebox has worked with third-party ASR solutions. With a high-quality, robust, in-house ASR solution, Voicebox can now double its language catalogue for the Embedded ASR platform over what it introduced in 2016. New languages will include Korean, Dutch, and Portuguese, and others. The expanded catalogue will also offer improved recognition for British, Australian, and Indian English dialects.
"The Talisman ASR enables us to deliver the entire end-to-end voice experience, from audio to understanding what the user said, to executing their request without the engineering overhead of integrating components from multiple suppliers," Dan Carter, vice president of speech at Voicebox, said in a statement.
Voicebox's Embedded ASR uses deep neural networks (DNN), which rely on vast quantities of collected data to generate probabilistic text outputs. Voicebox trains its speech models using a GPU compute cluster to include a wide range of speakers, ensuring their voice commands won’t be compromised due to differences in pronunciation or accent.