In order to give you the best web experience on our site, we use cookies. By continuing to use our website, you agree to the use of cookies.
Case Study
Creating more accurate voice-to-text transcriptions with Jabra headsets
Jabra’s mission is to improve people’s lives through the power of sound. We regularly put our products to the test, pushing boundaries to create the most advanced sound solutions that support contact centre agents in delivering quality customer service.
Industry
Location
Copenhagen, Denmark
Website
The Client
About Jabra
Jabra is a Danish brand specializing in audio equipment and more recently videoconference systems. It is owned by GN Audio, which is part of the Danish company, GN Group. Jabra engineers, manufactures, and markets wireless, true wireless, and corded headphones for consumers and business customers.
The Challenge
Improving transcription accuracy requires a multi-pronged approach.
In collaboration with NICE, Jabra conducted an experiment to establish the impact of Jabra headsets on speech analytics transcription output and accuracy vs non-Jabra headsets and earbuds in multiple environments, ranging from quiet to very noisy. The experiment was designed to simulate different contact centre scenarios.
To determine to what extent the quality of the headset/ microphone impacts the accuracy of voice-to-text transcription, and to what extent different background noise pollution impacts the transcription accuracy via the selected headset/microphone.
The Solution
A 1400-word test script (of which 831 words are considered ‘relevant’ to search/categorisation/algorithms) was read aloud by a contact centre agent a total of 15 times. Each time, one of five different headsets/microphones were used. The script was recorded across three different environments per headset – quiet background, mid-noise background and very noisy background.
All recordings were transcribed using ASR software, NICE ElevateAI (AI-driven ASR Technology with sentiment scoring and behavioural analysis.
Statistics
>94% individual word accuracy is an extremely high transcription score.
A ‘good’ transcription score for high- performance ASR software is >80%. Across a noisy environment, this is exceptional. Above 99% individual word accuracy, which the Jabra Engage 40 and the Jabra Engage 50 II achieved, is better than human-level transcription accuracy.
The Results
Jabra headsets support the most accurate transcriptions
Accuracy
All three Jabra headsets supported the most accurate transcriptions in a quiet environment, with >94% word accuracy and word relevancy accuracy. This equates to a Word Error Rate (WER) of <6%, which by industry standards is considered the gold standard. The generic headset failed to reach the required acuity to transcribe at all in a very noisy environment. In a controlled environment, it performed adequately, but it did not shine in any environment.
Consistency
All three Jabra headsets performed consistently across environments irrespective of the amount of noise pollution. The Apple AirPods performed well in a quiet environment but were highly impacted in a noisy environment.
Noise cancellation
All three Jabra headsets successfully eliminated background noise and maintained speaker acuity across multiple backgrounds interferences
Products used
Jabra
Engage 40
Jabra
Evolve2 Buds
Jabra
Evolve2 65 Flex
Featured Case Studies
