Recognition Accuracy

The Vestec Speech Engine delivers among the highest recognition accuracy in the industry. It is well suited for a wide variety of keywords, names, digits, numbers, dates, and yes/no speech grammars. In addition, the engine has been designed to operate in different audio channels – including, VoIP, cellular, and landline – in both “noisy” as well as “clean” environments.

Generally speaking, without grammar “tuning”, the speech engine delivers recognition accuracy in the 90% range for native speakers and in the 80% range for non-native speakers. Grammar “tuning” by the application developer typically improves recognition accuracy by 5-10%.

Accuracy Study

Vestec regularly benchmarks recognition accuracy against leading competitive offerings and has consistently demonstrated comparable performance. A recent accuracy study involved tests with third-party data consisting of audio recordings of the five hundred most commonly spoken words. The recordings represented a diverse group of male and female voices in different background and channel environments. For both native and non-native speakers, Vestec ASR outperformed the sample group of speech engines by nearly 3% points in recognition accuracy.

RECOGNITION ACCURACY	Vestec Speech Engine	Average Improvement over Competition
Keywords - Native Speakers	In the 90% range	+3.2%
Keywords - Non-native Speakers	In the 80% range	+3.0%