A hidden Markov Model (hmm) based speaker identification system using mobile phone database of North Atlantic Treaty Organization (nato) words

Experimental Results:
One hundred speakers have been selected to conduct the experiment, out of 
these 70 speakers have been used to train the system and then comparison of every speaker has been made 
with each other speaker. Also 30 different speakers have been taken as a test data and compared with all 70
trained speakers. The comparison with correct recognition and the false acceptance are shown in Table 1.
Agrawal et al.
Proceedings of Meetings on Acoustics, Vol. 19, 060019 (2013) Page 3

---

Number
Of Speakers
Utterances
Correct Recognition 
(%)
False Acceptance (%)
70(Comparison
with same speakers)
32200
74.28
25.72
30(Comparison with different 
speakers)
13800
0
26.66
Table-1
Threshold value has been taken as 65 on the basis of observations of the whole experiment.
The speakers with largest and smallest accuracies from the whole experiment are shown in Table 2.
Nearest
Speakers
Accuracy (%)
Farthest
Speakers
Accuracy (%)
Speaker
24&45
99.06
Speaker
1&61
20.55
Speaker
55&60
98.89
Speaker
52&69
32.52
Table-2
On the basis of comparison, the speakers having largest and smallest accuracies have been considered as 
nearest and farthest speakers.
Recognition Performance Based on Different Utterances of the Speakers:
The system
was trained from different numbers of utterances of the speakers. These utterances in slots of first 5,
first 10 and first 15 were used as training samples. The remaining 5 utterances were used as test samples.
The results obtained from this analysis are shown in table3.
Speakers
Trained
Utterances
Test
Utterances
Accuracy (% Recognition)
100
Speakers
1
st
To 5
th
16
th
To 20
th
67.90
1
st
To 10
th
16
th
To 20
th
75.30
1
st
To 15
th
16
th
To 20
th
79.06