Speech Recognition for Voice-Enabled Devices: When Accuracy Matters
We’ve been challenging the existing status quo of voice-capture speech recognition technology in the marketplace for several years now. The performance of existing OEM solutions for voice-enabled devices just doesn’t cut it anymore. They lack range, do not operate optimally in loud or noisy environments or spaces with other competing noise sources, and their accuracy in capturing voice commands is subpar.
Recently, a public safety technology company asked us to test the transcription capabilities of our EveryWord™ Ultra Far-field Voice solution against their current voice capture solution. For this client, accuracy in capturing voice commands was key.
In a side-by-side WRR (word recognition rate) using an ASR evaluation tool, the results were clear: EveryWord outperformed the client’s current solution with a 98.27% accuracy rate vs. an 80.53% rate.
That is a remarkable difference.
This is not the first time that EveryWord has proved its exceptional human-to-human or human-to-machine speech recognition performance when we go head-to-head with the “status quo.” Not only does EveryWord capture voice commands with superior clarity and accuracy at 3X the distance (up to 9 Meters) of standard far-field technology, around corners and obstacles, and in noisy and reverberative environments, but it also identifies and suppresses speech from other single-point noise sources.
If your team would like to test our solution for yourself, you can order our development kit here.