Fall 2022 Product Release: Improved Speaker Recognition and Speech to Text Accuracy
October 13, 2022
By Pavel Jiřík in Blog
Leaving summer behind, Phonexia’s research and development team has advanced the capabilities of speaker identification and speech recognition technologies once again.
Let’s see what Phonexia’s latest Fall 2022 Product Release has to offer for your innovative projects.
The Most Accurate Speaker Identification Yet
Phonexia has been leading the field of voice biometrics for many years now, pushing the technology's limits further every year.
Therefore, it is with great excitement that we announce our brand-new fifth generation of Speaker Identification technology.
This generation has the code name XL5 and has been rewritten from top to bottom to reflect the latest breakthroughs in the science of voice biometrics and deep neural network approaches.
As a result, its speaker identification accuracy has increased by one percentage point, which is up to a 30% relative improvement in the accuracy.
All that while being backward compatible with the voiceprints generated by our previous Speaker Identification generation, XL4.
This means our customers can seamlessly update to the fifth generation of Speaker Identification to enjoy a significant increase in speaker recognition accuracy.
Speech to Text Accuracy Improvements
We have upgraded seven more languages to our latest sixth generation of Speech to Text technology, powered by advanced deep neural networks and innovative speech recognition enhancements:
Arabic (Gulf) Speech to Text
Upgraded from the fourth generation, it can transcribe Arabic with a word accuracy of up to 58%.
Chinese (Mandarin) Speech to Text
Trained from scratch, this Speech to Text model transcribes Mandarin Chinese with an impressive word accuracy of up to 93%.
Dutch Speech to Text
Upgraded from the fifth generation, Dutch is transcribed with a word accuracy of up to 86%.
German Speech to Text
Upgraded from the fourth generation, this Speech to Text model transcribes German with a word accuracy of up to 90%.
Italian Speech to Text
Being a significant update from our old third Speech to Text generation, Italian can now be transcribed with a word accuracy of up to 78%.
Polish Speech to Text
Upgraded from the fifth generation, this Speech to Text model achieves a word accuracy of up to 85%.
Russian Speech to Text
Upgraded from the fifth generation, Russian can be transcribed with up to 90% word accuracy.
New Speech to Text Language – Georgian
In addition to the above upgrades, we have also expanded our list of Speech to Text languages.
The new Georgian Speech to Text language is a brand-new model that achieves a word accuracy of up to 66% on our evaluation datasets.
In total, Phonexia now offers 20 Speech to Text languages, all using our most advanced sixth generation of Speech to Text technology.
Updated Speech to Text Dictionary
To complete the list of Speech to Text accuracy improvements, it is also important to mention that the word dictionary for Czech, Dutch, Georgian, German, and Slovak has been updated to include the most recent words, such as COVID.
The update makes the speech transcription more accurate for the aforementioned languages, and our team plans to update the dictionary on a regular basis to reflect the fast-paced nature of our world.
Easier Deployment of Phonexia Speech Platform
Phonexia Speech Platform is now easier to deploy as it can configure itself automatically based on detected hardware, ensuring the best possible usage of computing power.
To make the deployment even easier, the Phonexia Speech Platform is now available in the Docker Hub public repository. From now on, you can build large-scale containerized applications much more quickly.
The future holds many variables, for sure. Nevertheless, we can already share with you that our researchers and developers are working on two new Speech to Text languages – Bengali and Kazakh.
We are also working relentlessly on the new versions of our other cutting-edge products, Phonexia Voice Inspector and Phonexia Orbis Fraud Detect.
Plus, our voice biometrics experts are exploring new innovative ways to push voice recognition technology even further.
If you are interested in more detailed product release notes, you can find them on our Partner Portal.