University Tests VoiceVantage's Biometrics Voice Verification using Two Sets of Identical Twins

date

Study conducted at Ontario Tech University (UOIT), last updated 2024. Published here October 2026.

Identical twin sisters smiling side by side in a bright kitchen (stock photo, not the twins in the study)

Gauge of the VoiceCheck imposter score, out of 100: on the favorite color phrase the imposter averages stayed between 0 and 4.5

Identical Twin Test 2 · Ontario Tech University (UOIT) capstone study

Can one identical twin pass as the other?

Two sets of identical twins enrolled their own voices, then each twin tried to pass as the other. The study tested VoiceVantage VoiceCheck™ on two microphones and over VoIP, and the demo tools of two other voice biometric products.

2

sets of identical twins

4 × 4

phrases × renditions

2 + VoIP

microphones and a VoIP line

3

products tested

Banner photo: stock image, not the twins in the study.

Key result

Exclusive phrase: near zero

When the imposter twin had to say the other twin’s favorite color, the VoiceCheck™ imposter averages stayed between 0 and 4.5 out of 100.

Why it works

Why VoiceCheck™ uses personal phrases

On “My Voice Is My Password” and birth city, true for both twins, the imposter averaged 37 to 54 out of 100 on the microphones, and 68 to 73 over VoIP. On a phrase only one twin can say truthfully, it stays near zero.

Other products

Other products: first attempt

On the Microsoft-Azure demo tool and the Verint-VoiceVault ViGo 2 app, both sets of twins passed as each other on their very first attempt. Microsoft has since retired Azure Speaker Recognition (September 2025).

Abstract

 

Voice biometrics has been used successfully to determine and verify speaker identification for years yet differentiating identical twins has always been a challenge.

This capstone project presents the in-depth details of the biometric voice verification process and the usage of identical twin test analysis to validate its efficiency. Recorded voice samples from identical twins have been verified using VoiceVantage’s SDK and other voice biometric tools from Verint-VoiceVault and Microsoft-Azure. Results were analyzed statistically to determine the accuracy of the VoiceVantage tool in detecting imposter attacks.

Voice samples were collected in a controlled environment from identical twins and verified against one another to obtain their respective comparison scores. The scores have been analyzed and summarized to derive the outcome of the test and relate the scores to the nature of the attempts on each passphrase and its properties.

In addition to VoiceVantage, details behind the voice verification process and the history behind various prominent voice biometric authentication products in the market such as VoiceVantage VoiceCheck™, Microsoft-Azure speaker verification and Verint-VoiceVault have also been explored. Characteristics of human voice, speech analysis process and voice verification process are analyzed in detail as part of this project. The above-mentioned products have been subjected to the Twin Tests as part of this capstone project as well.

Capstone project report

Performance Evaluation of Biometric Voice Verification Tools with Identical Twins

Faculty

Faculty of Business and Information Technology

University

Ontario Tech University (UOIT)

Author

Tamilarasan Muthu

Sponsor

THC Technologies Corporation (see the Disclaimer)

Tested

VoiceVantage VoiceCheck™ (32-bit and 64-bit), Microsoft-Azure, Verint-VoiceVault

Not tested

Microsoft’s Nuance (no demo tools)

The test set-up at a glance

 

Twins

Two sets of identical twins: Twins 1 and Twins 2

Phrases

4 phrases, 4 renditions each, as self and as imposter

Devices

Logitech H600 headset, Blue Yeti Pro microphone, Cisco SPA 303 VoIP phones

VoiceCheck™

32-bit and 64-bit tools, similarity score from 0 to 100

Other products

Microsoft-Azure web demo, Verint-VoiceVault ViGo 2 iOS app

Results: average scores by phrase

 

Phrase 1: First and last name (aliases)

Candidate = (A + B) / 2, imposter = (A′ + B′) / 2, average scores out of 100

  Candidate (own voice)

  Imposter (the other twin)

Cand.

Imp.

Twins 1

32-bit · Logitech

 
 
 
 

57.0

1.5

32-bit · Blue

 
 
 
 

75.0

8.6

64-bit · Logitech

 
 
 
 

67.0

0.2

64-bit · Blue

 
 
 
 

69.0

7.3

Twins 2

32-bit · Logitech

 
 
 
 

68.5

1.4

32-bit · Blue

 
 
 
 

63.0

4.5

64-bit · Logitech

 
 
 
 

67.0

0.4

64-bit · Blue

 
 
 
 

64.0

1.4

 

0

25

50

75

100

 

Phrase 2: “My Voice Is My Password”

Candidate = (A + B) / 2, imposter = (A′ + B′) / 2, average scores out of 100

  Candidate (own voice)

  Imposter (the other twin)

Cand.

Imp.

Twins 1

32-bit · Logitech

 
 
 
 

74.5

49.3

32-bit · Blue

 
 
 
 

76.0

40.1

64-bit · Logitech

 
 
 
 

76.0

51.9

64-bit · Blue

 
 
 
 

74.5

39.2

Twins 2

32-bit · Logitech

 
 
 
 

72.5

44.5

32-bit · Blue

 
 
 
 

75.5

49.8

64-bit · Logitech

 
 
 
 

70.0

40.8

64-bit · Blue

 
 
 
 

67.5

42.2

 

0

25

50

75

100

 

Phrase 3: Birth city and state

Candidate = (A + B) / 2, imposter = (A′ + B′) / 2, average scores out of 100

  Candidate (own voice)

  Imposter (the other twin)

Cand.

Imp.

Twins 1

32-bit · Logitech

 
 
 
 

71.0

42.4

32-bit · Blue

 
 
 
 

69.5

37.3

64-bit · Logitech

 
 
 
 

76.0

44.3

64-bit · Blue

 
 
 
 

71.0

53.8

Twins 2

32-bit · Logitech

 
 
 
 

71.0

43.1

32-bit · Blue

 
 
 
 

68.5

45.7

64-bit · Logitech

 
 
 
 

64.5

40.2

64-bit · Blue

 
 
 
 

66.5

41.8

 

0

25

50

75

100

 

Phrase 4: Favorite color

Candidate = (A + B) / 2, imposter = (A′ + B′) / 2, average scores out of 100

  Candidate (own voice)

  Imposter (the other twin)

Cand.

Imp.

Twins 1

32-bit · Logitech

 
 
 
 

73.0

0.0

32-bit · Blue

 
 
 
 

74.0

0.0

64-bit · Logitech

 
 
 
 

73.0

0.0

64-bit · Blue

 
 
 
 

75.5

0.0

Twins 2

32-bit · Logitech

 
 
 
 

75.0

4.5

32-bit · Blue

 
 
 
 

68.0

0.0

64-bit · Logitech

 
 
 
 

74.0

1.4

64-bit · Blue

 
 
 
 

67.5

0.3

 

0

25

50

75

100

 

VoIP: “My Voice Is My Password”

Over VoIP phones, average scores out of 100

  Candidate (own voice)

  Imposter (the other twin)

Cand.

Imp.

Twins 1

32/64-bit · VoIP

 
 
 
 

81.0

73.2

Twins 2

32/64-bit · VoIP

 
 
 
 

79.0

68.4

 

0

25

50

75

100

 

Want the detail?

Read the full technical report

How voice verification works, the recording equipment and every score table behind these charts.

Read the full technical report

Twin Tests with Microsoft-Azure Speaker Verification

 

To subject Microsoft-Azure’s speaker verification tool to twin testing, the web-based demo tool has been chosen, by using our candidates Twins 1 and Twins 2. Out of all the built-in phrases available, the phrase “My voice is stronger than passwords” has been selected since it is the closest match to the phrases used for all the products tested for this capstone project.

Both Twins 1 and Twins 2 were able to successfully complete the enrollment process and able to successfully authenticate against each other with confidence score ‘high’ on their very first attempts.

Microsoft has since retired Azure Speaker Recognition (September 2025).

Enrollment: done by both twins

Imposter attack: accepted, confidence ‘high’

 

 

 

Speaker verification demo tool

Figure 29 of the UOIT report: Microsoft’s Azure speaker verification demo tool

Swipe sideways to see all of it →

Figure 29. Microsoft’s Azure speaker verification demo tool

Figure 30 of the UOIT report: Microsoft’s Azure Speaker verification - enrollment

Swipe sideways to see all of it →

Figure 30. Microsoft’s Azure Speaker verification - enrollment

Figure 31 of the UOIT report: Microsoft’s Azure Speaker Verification – imposter attack result

Swipe sideways to see all of it →

Figure 31. Microsoft’s Azure Speaker Verification – imposter attack result

Twin Tests with Verint-VoiceVault

 

VoiceVault’s iOS app called ‘ViGo 2’ has been used to test with the twins. Both Twins 1 and Twins 2 were able to successfully register, authenticate by using the phrase ‘At VoiceVault, my voice is my password’ which is the closest match to the phrases used for testing the other products.

Twins 1 and Twins 2 were able to successfully impersonate each other on their very first attempt without any difficulty.

A woman giving a voice command to her phone

Figure 32 of the UOIT report: ViGo2 – iOS app – user profiles

Figure 32. ViGo2 – iOS app – user profiles (names replaced with labels)

Figure 33 of the UOIT report: ViGo2 – iOS app – imposter attack result

Figure 33. ViGo2 – iOS app – imposter attack result (names replaced with labels)

Conclusion

 

“Voice is just not mere words, it carries the emotional characteristics of the person at the moment of recording.”

From the report’s conclusion

On analyzing the comparison scores generated by the VoiceVantage VoiceCheck™ tool, it is clearly evident that the possibility of successfully spoofing among the twins is not possible provided the phrases are exclusively specific to the candidate. The comparison score of the imposter went above 60% but below 75% only for phrases 2 and 3 which are “My voice is my password” and “city/state of birth” respectively using conventional recording devices. Over VoIP, the imposter’s scores are between 60% to 77% on average consistently.

In both cases, the phrases are factually ‘true’ for the imposter. On average the comparison score is fairly high when the imposter, no need to lie or to make an effort to impersonate his counterpart. This is clearly proving that voice is just not mere words, it carries the emotional characteristics of the person at the moment of recording. Voice tone and the difference in the psychological intent of telling the actual truth and trying to impersonate are really causing a huge difference in the voice samples.

The scores for the phrase ‘favorite color’ which is factually different for each of the twins and the imposter score for this phrase went absolute minimum when they must really try to spoof each other. This proves that choosing the phrase and its degree of exclusiveness to the individual is also a key factor in effective and efficient authentication.

The sensitivity of the voiceprints generated by VoiceVantage VoiceCheck™’s algorithmic technology to all those above-mentioned factors and properties of voice is very commendable and the results clearly prove that it is highly resistant to any such imposter attacks. Also, the performance of both the 32-bit and 64-bit tools are very similar.

We attribute the negligible differences in similarity scores between the two microphones to the hardware. We are certain that the collection environment and input quality were the same for both the devices. The test result shows that using the recording device with ideal settings/configuration will make a significant difference in determining the 'real' threshold and reduce the false acceptance rate.

Even though the nature of the tools is different (iOS app, web-based API tool), the imposter attacks were successful effortlessly on the other products. Considering the imposter attacks’ success rate on the products we tested and the range of similarity scores of the samples collected for VoiceVantage, it is clear that VoiceVantage is above par in efficient voice verification provided the phrases are uniquely related to the candidates.

Disclaimer

 

Please note that this research work was sponsored by THC Technologies Corporation. Again please note: The tools used for testing the THC VoiceVantage VoiceCheck SDK were provided by THC, and according to THC, the tools were in no way modified to provide any strategic advantage over its competitors. Said tools were designed only to aid in simplifying the sample collection and analysis process.

The research findings and reviews expressed in this documentation are truly unbiased and strictly based on the statistics gathered during the project with no intention to disrepute any of the products/vendors mentioned in the documentation.

Identical Twin Test 1

Read the first identical twin study

UOIT researchers had identical twins imitate each other on two microphones and scored every attempt with VoiceCheck™.