February 2022 Summaries

The development of voice technology has reached a critical juncture where understanding the emotional nuances behind human speech is becoming increasingly important. Current voice assistants often struggle to recognize emotions, which can lead to misinterpretation and a breakdown in dialogue. Emotion recognition is essential for creating an empathetic and affective conversation between humans and machines. Research is being conducted by companies such as Speechmatics to improve emotion recognition accuracy, including the use of self-supervised learning and exploring diverse data sets like TV broadcasts and phone calls with less human-labeled data. By analyzing real-world scenarios with varying levels of noise and emotions, these studies aim to better understand how emotions impact speech-to-text technology and develop more accurate and inclusive systems.

The voice technology industry has made significant progress despite the ongoing pandemic, with a growing investment in deep tech and academia. The annual Voice Report highlights key findings on AI bias, which affects many users, particularly those from diverse backgrounds. Despite this, there is an increasing appetite for real change among investors and developers. Self-supervised learning is seen as a critical solution to address biases, and its adoption is expected to play a crucial role in making tangible progress. The report also notes the growing demand for specific languages, such as Cantonese and Brazilian-Portuguese, and the importance of accuracy in voice technology, with breakthroughs in non-American English languages contributing to reduced accuracy gaps among different demographics.

Speechmatics has had an incredible year in 2021, expanding its team and improving its core speech recognition engine to achieve new accuracy levels. The company launched its Autonomous Speech Recognition technology, which garnered global press coverage and was named the number one speech recognition company using ML and AI by A.I. Magazine. Speechmatics also introduced self-supervised learning into its training, resulting in extraordinary accuracy improvements. Additionally, the company made various product updates, including new features such as language improvements, usage API access, and improved speaker diarization. These advancements have positioned Speechmatics for continued growth and success in 2022.