any language, without bias to accents, regional dialects
or speaking conditions. Welcome to our evolving world
VoicEncode’s engine leverages our innovative speech2vec algorithm and uses the most sophisticated machine learning technology to detect and understand accents under various speaking conditions and pronunciation differences. The algorithm is trained with minimum data and when combined with existing ASR systems, dramatically improves the performance at minimal cost
Our core technology is powered by a highly sophisticated efficient speech encoding algorithm that extracts the vocal configuration of spoken words to be analyzed via spectrograms. It then normalizes the extracted features to consistent representation, enabling easy integration with machine learning tools, e.g. Random Forest. This unique technology enables us to boost traditional ASR systems through an end-to-end training, without the need for HMMs, DNNs and phonetic dictionaries traditionally being used in the industry
recognition flow through our 3-stage process:
During this stage, send us recordings and the engine’s (e.g. Google Cloud Speech API) results, including confidence scores
Comparing the engine’s results with the actual human-transcriptions will set your accuracy baseline
Set your performance goals (min. true recognition and/or max. false recognition rates) and get calibrated thresholds for the confidence scores
All speech recognition solutions use machine learning tools that keep learning and improving. Their main drawbacks are a need for vast amount of data and a strong bias toward majority groups
Our unique technology requires considerably less data and learns YOUR data much faster
At the end of the Measure stage, keep sending us recordings for free to get speech recognition tailored to your data
When enough data is gathered for initial calibration of our models, start paying only for cases that fall below the calibrated thresholds
Monitor your improving performance through our periodic analysis reviews to make sure you’re getting the desired experience
Let’s bring great speech recognition to everyone. Together!
Our platform supplies you with speech-metrics that state the accuracy for all of your users
The discrimination-factor that represents the bias towards certain groups is one of the figures we aim to reduce for the benefit of your users
VoicEncode’s technology was not developed for a specific language and can improve accuracy for many languages globally
In addition, the (non-deep) machine learning tools we are using allow us to deliver you the same voice experience for various speaking styles (e.g. accents and speech-impairment) and diverse speaking conditions
We developed our architecture as a RESTful API to give you an easy and low-cost integration
You can choose whether you want to share your data with others and benefit from their data, or keep your data private and secure
Contact us for a live demo
No commitment evaluation
See how fast your accuracy can improve