audEERING®, a leader in voice analytics technologies, has embarked on a strategic journey, offering both open-source models and commercial solutions. This blog series will first explore the innovative open-source models for age & biological sex detection, and expression analysis, which are accessible to a wide audience and serve as a springboard to the more sophisticated capabilities found in audEERING’s commercial product, devAIce®.
Open-source models for research and academic use
audEERING® has released, for academic use and research only, an open-source model on Hugging Face for predicting age and biological sex from voice. A groundbreaking tool based on data from three public databases. Presented at the 15th ITG Conference on Speech Communication at RWTH Aachen University, this model is a significant step toward democratizing voice-driven technologies.
The connection between age and biological sex
The model distinguishes between two sex classes and one age-related third class: female, male, and children – the latter are not distinguished by biological sex in the voice as adults are.
The biological sex model outputs 2 binary classes (male and female) + the age in years in the age model. The connection between these models lies in the child’s voice having different acoustic structures, making a gender-specific distinction impossible. Therefore, if a voice is identified as a child’s voice, it receives the model label “child” and no sex assignment. The ages of adults and children can be identified accurately with a deviation of just a few years. This approach broadens the model’s applicability and aligns with the growing demand for inclusive technology.
With over 12,000 downloads in March 2024 alone and a total download of almost 100,000 (for two model versions) in less than a year, the impact of the biological age-gender model is evident and reflects audEERING’s commitment to transparency and collaboration in the digital age. The open-source model supports researchers and drives invocation worldwide.
The devAIce® commercial suite – easy to implement
For commercial users, audEERING® offers a commercial age model with the highest accuracy and validated robustness, fully integrated into the devAIce® SDK and Web API products. The devAIce® commercial suite is tailored for organizations that require accurate age and biological sex analysis.
The comparison in the figure shows the open source and 2 generations of commercial models. We show that open source follows the quality standard that we have set with our commercial models and is constantly improving.
Explanation: CCC – coefficient indicating the correlation between the actual and predicted age; the higher, the better. The mean average error in years divided by 10, the lower the better. The result of the public model is on average 10.9 years off the actual age.
Prepaid packages with all AI modules – with a limited offer
Using the devAIce® Web API you can choose between two prepaid options with a limited offer that includes all devAIce® packages.
The Speaker module is equipped with age recognition and so-called perceived gender recognition.
For more information, browse our redesigned website and get in touch with us.