Study on voice recognition utilizing bone-conducted voice

movedearΤεχνίτη Νοημοσύνη και Ρομποτική

17 Νοε 2013 (πριν από 3 χρόνια και 4 μήνες)

51 εμφανίσεις

Study on voice recognition utilizing bone-conducted voice
Yoshiyasu Yamada
(DENSO Corporation, yyamada@rlab.denso.co.jp)

Yoshimasa
Hijikata
(DENSO Corporation, yoshimasa_hijikata@denso.co.jp)

Nobuaki Kawahara

(DENSO Corporation, nkawaha@rlab.denso.co.jp)
Yoichi Fujisaka
(National Institute of Advance Science and Technology, y-fujisaka@aist.go.jp)
Seiji Nakagawa
(National Institute of Advance Science and Technology, s-nakagawa@aist.go.jp)

Keywords:
bone-conducted voice, voice recognition, piezoelectric bimorph, sound quality evaluation, ubiquitous

Voice Recognition using bone-conducted voice has
been investigated. The sound quality of bone-conducted
voices detected at several points on human head was
estimated in order to examine the suitability for voice
recognition and voice communication. LPC cepstrum
distance (LCD), which indicates the vector distance
between air-conducted voice and bone-conducted voice,
and Speech Transmission Index (STI), which indicates
the voice clarity for human ears, are used for the
estimation. According to the experimental results, it is
clarified that forehead and cheek are suitable for both
voice recognition and voice communication.
In order to obtain a high voice recognition ratio with
bone-conducted voice, which is comparable to that with
air-conducted voice, a novel bone-conduction microphone
was fabricated. The microphone has an inclined
frequency response so that the obtained bone-conducted
voice through this microphone has similar characteristics
to those of air-conducted voice. Figure 1 shows the
schematic diagram of the microphone. A pair of
piezoelectric bimorphs with weights at their edges is
stacked, and is fixed with a M0.6 screw. The size of
bimorphs and weights are precisely simulated by FEM
technique, so as to realize the ideal frequency response.
The frequency response of the fabricated microphone is
shown in Fig.2. The resonance point is located around 4
kHz, and the sensitivity gradually decreased toward lower
frequency range.
Bone-conducted voices were obtained at human
forehead using this microphone, and their frequency
responses were examined. Figure 3 shows the frequency
response of air-conducted voice, bone-conducted voice
through an acceleration sensor with flat sensitivity, and
bone-conducted voice through the novel microphone,
respectively. The results indicate that the novel
microphone can obtain bone-conducted voices with
similar frequency responses to that of air-conducted
voice.
Voice recognition experiments were carried out using
the microphone. Commercially available car-navigation
system was used for the experiment. The obtained
recognition ratios varied from 30% to 95% with each
subject, which were not appropriate for actual use.
To overcome this problem, Spectral Subtraction
method was applied to improve Signal to Noise ratio
(S/N). The results show that high recognition ratios
over 80% are achieved for all subjects at S/Ns higher than
30dB. This indicates that voice recognition using
bone-conducted voice is feasible enough for actual
products such as cellular phone, PDA, wearable
computers, and so on.
10.2mm
weight
piezoelectric bimorph
outer case
board
10.2mm
weight
piezoelectric bimorph
outer case
board

Fig.1 Schematic diagram of the prototype of bone-conduction
microphone
      










4FOTJUJWJUZ N7NT


'SFRVFODZ L)[


Fig.2 Frequency response of the bone-conduction microphone
      






BJSDPOEVDUFEWPJDF
CPOFDPOEVDUFEWPJDF QSFTFOUNJDSPQIPOF

CPOFDPOEVDUFEWPJDF BDDFMFSBUJPOTFOTPS



(BJO E# BV

'SFRVFODZ L)[

TVCKFDU#

Fig.3 Frequency response of air-conducted voice and
bone-conducted voice