About the Study

Many factors affect the performance of ASR systems, including background noise and reverberation, echo, and distance from the speaker.
‍
The following study discusses these challenges and examines solutions to produce more accurate, intelligible ASR performance for voice and speech recognition devices. It shows Kardome’s speech recognition enhancement technology outperforming standard industry algorithms.

We measure Wake Word False Rejection Rates and Response Accuracy Rates by studying ASR performance in various environments, from the quietest to the noisiest.

In the study we look at the following Performance Indicators:

• Response Accuracy Rate (RAR) — the percentage of times the device correctly responds.
• Wake Word Detection Delay (DD) — the time it takes for a system to recognize the wake word and be ready to accept a further command. To have a real-time and enjoyable user experience, the WWDD should be as small as possible.
• Wake Word False Alarm Rate (FAR) — the percentage of times a user activates the device without the wake word being spoken.
• Wake Word False Rejection Rate (FRR) — the percentage of times the device does not respond to the wake-word. For example, an FRR of 10% shows a 90% success rate.