Frequently Asked Questions (FAQ)
A collection of common questions and concerns related to voice biometric technology
Accuracy and Reliability
The accuracy of voiceprints is comparable to that of fingerprints and other commonly used biometrics. VBG regularly tunes production verification systems to 97-99% accuracy levels, depending on the usage environment and quality of the microphone/device among other factors. The VMM-3 voice biometric engine uses both physiological and behavioral characteristics of your voice to create a unique voiceprint. And since a majority of these characteristics tend to be consistent over time, they can be accurately measured under varying conditions.
Yes. The VMM-3 voice biometric engine models many different vocal features. Our scoring system models and scores every feature individually and combines them into a composite score. So, if there are small changes to a few features, such as might occur with a cold or allergies, the overall score can still be accurate (and the user properly verified).
This is extremely unlikely. A person’s voice is unique to them. And while some professional mimics and comedians might seem to sound like someone else, the VMM-3 voice biometric engine measures and scores vocal characteristics that a human being cannot hear (or recreate). Further, many client applications use random prompting techniques to help eliminate prior knowledge of what will be requested.
In most cases VBG's audio analysis and filtering tools can handle background noise, allowing the VMM-3 voice biometric engine to process speech samples without issue. When noise levels are too high however, the VBG System responds back to the client application so that it can re-prompt the user accordingly. As a general rule, users should be able to speak comfortably from a location with typical noise. Users should avoid loud and noisy areas that are crowded (sporting events, bars, concerts, etc.), as there will be too much audio information for the VBG System to reliably process.
Yes. People typically have access to several different devices, so it is impractical to require users to only use one. The VBG System was designed to accept audio from any application source (IVR system, cell phones, PC or device microphone, etc.). And while it is true that the best results typically occur when a user enrolls and verifies using the same device, VBG's tuning process uses sophisticated statistical models to help normalize the variations that occur between different audio sources.
VBG takes the collection of personally identifiable information (PII) very seriously. We only collect speech samples (source WAV audio or media streams) which we use to process into voiceprints, or to compare to stored voiceprints during verify and identify requests.
For clients using our optional Automatic Speech Recognition (ASR) features, we may also collect content descriptions for the speech samples that are sent to us. And, if our clients and partners are using a VBG partner's integrated IVR system, the use of outbound calling requires that telephone numbers be sent to us.
Personal data can only be sent to the system via VBG's secure APIs, which require an account code and API access token. We further require all clients and partners to use HTTPS to transmit data to us. Once data is received in the VBG System, it is either discarded from memory or written to our secure database. Any personally identifiable information (PII) that is stored in VBG's database is encrypted using AES-256 encryption under FIPS 140-2 guidance.
The VMM-3 voice biometric engine and VBG Platform are fully developed and maintained by VBG's team of voice biometric engineers, system architects, researchers, and developers. Our core team has years of experience in voice biometric engineering, digital signal processing, and software development.
The VMM-3 engine contains a number of different machine learning algorithms, such as GMM, SVM, and i-Vector. The VMM-3 engine also includes newer "deep learning" techniques, such as convolutional networks and bi-directional LSTM networks.
Different algorithms are better suited to different use cases, and also depend on how much training data is available to tune the system. VBG works with our clients and partners to configure the VMM-3 engine with the optimal algorithm for their particular circumstances.
The VMM-3 engine supports standard WAV audio files with headers. Acceptable forms include: PCM, u-Law, and a-Law. Your audio files should be mono (single channel) and only include speech from one speaker. Files can be 8-bit or 16-bit and either 8kHz or 16kHz.
NOTE: the vast majority of available speech information can be determined from 16-bit, 16kHz samples, so anything more than this will not really buy you much in the way of accuracy. Hence, we do not accept higher resolution files. There is also storage space to consider when scaling and managing very large deployments, so using optimal file formats is critical in these situations.
VBG does not currently have a formal partner program. However, we do have many partners globally and have many of the expected components of a partner program, such as: partner pricing discounts, partner resale agreements, partner technology training and support documentation, marketing support documents, meeting preparation and presentation assistance, etc.
The most successful VBG partners actively use our technology, either to develop custom applications for their clients, or to embed and host within their own products and services, or to combine with their own products and services for resale. VBG currently has no pure "resale only" partners.
Clients using VBG's Shared Hosting System can choose from several different subscription plans, supporting volumes up to 2MM transactions per month. If volume requirements exceed 2MM transactions per month, VBG may require private instances within our Hosting System. It's important to note that VBG offers volume discounts; the higher the monthly transaction volume commitment, the lower the per-transaction price.
Yes! We provide 60-day free trials of the VBG Platform. Free trials are hosted in our datacenter using a deployment that is identical to the one used by our production clients. So, you'll be assured that your intergration work, system performance, and results are consistent. You can also test optional telephony and IVR services, with both inbound and outbound calling, often with no additional charges.
Yes. If you do not want to want you use the VBG Hosting System, you can deploy on-premises. VBG currently has private deployments in over a dozen different countries. And, if you prefer a public cloud such as Alibaba Cloud, Amazon AWS, Google Cloud, IBM Cloud or Microsoft Azure - we're currently maintaining private production deployments for clients in each of these environments too.
Yes. In fact, VBG does not have its own IVR system either. We instead have built plug-ins to support numerous IVR platforms. Thus, clients can directly incorporate standard enrollment and verification dialogs into their applications without the need for any IVR programming. VBG currently supports IVR integrations with Alvaria, Convergys, Intrado, PlumVoice, Telnyx, and Twilio. Other platforms are possible as well, so please inquire if interested.
It is also important to note that the VBG Platform can accept audio from almost any source. We have clients providing us with audio from IVR systems, call centers, mobile applications, WebRTC, Bluetooth speakers, traditional microphones, medical devices, and IoT devices. As long as you can provide speech samples to us via one of our supported audio file formats (or streaming), we should be able to process voice biometric requests for you.
Due to VBG's comprehensive security protocols, it is extremely unlikely that any voiceprint will ever be stolen or hijacked. However, if a voiceprint were to find its way outside of the VBG Platform, absolutely nothing would happen! There are a couple important reasons why.
First, VBG's voiceprints are essentially mathematical models of how a person speaks. They are not files that can be played or turned into intelligible speech. Plus, VBG uses a proprietary format, so only VBG's system can interpret and use them. And second, voice biometric systems work by submitting speech samples to them, extracting features, and then comparing these extracted features to a reference voiceprint. Simply put, a voiceprint cannot be submitted directly to VBG's system, nor to any competitor's system.