COMPACT HELIUM SPEECH UNSCRAMBLER USING CHARGE TRANSFER DEVICES

Dr M.A. Jack* Dr A.D. Milne, Dr L.E. V i r r . A

ABSTRACT

This paper describes the p r i nc ip l e , design and performance of a compact helium speech unscrambler using charge t ransfer devices. Helium speech is the term commonly used for the d is tor ted speech uttered by deep-sea divers breathing in an atmosphere which consists of a mixture of helium and oxygen gases. The unscrambler considered here operates in p i tch synchronism with the input helium speech waveform and uses time-expansion signal processing to achieve a considerable improvement in the i n t e l l i g i b i l i t y of the speech s igna l .

INTRODUCTION

Deep sea divers breath a mixture of helium and oxygen gases in order to overcome the tox ic and narcot ic problems produced when breathing high pressure a i r . However, speech produced in such a high pressure helium/ oxygen mixture suffers a unique and severe form of d i s to r t i on known as 'helium speech 1. The main features of the helium speech d is to r t i on can be summarised as:

* A l inear s h i f t in vowel formant frequencies {1,2} * Minimal change in fundamental frequency (p i tch) {3} * A non-l inear s h i f t in low frequency formants {2} * Minimal change in formant bandwidths {4} * Reduction of unvoiced sound in tens i t i es {4}

Present helium speech "unscrambler" designs take into account only the f i r s t two of these properties {4 ,5 } . In these systems, the d i s t o r t ­ions present in the helium speech waveform are reduced by signal proc­essing techniques based on spectrum compression - by means of waveform time expansion - in p i tch synchronism with each p i tch period {5} . Such signal processing requires electronic storage and current systems gen­e ra l l y employ d i g i t a l memory for th is funct ion.

The unscrambler design considered here uses s imi lar time expansion techniques but employs analogue charge t ransfer devices fo r wave storage with CMOS d i g i t a l c i r c u i t r y fo r control functions and potent ia l rea l isa t ion as a single integrated c i r c u i t . Min ia tur isat ion of the system would permit the unscrambler to be carr ied by the d iver , leading to improvements in through-water, d iver - to -d iver communication systems.

Dr M.A. Jack, E.E. Dept., University of Edinburgh, EDINBURGH. Dr A.D. Milne, Wolfson Microelectronics I n s t i t u t e , EDINBURGH. Dr L.E. V i r r , Admiralty Marine Tech. Est . , PORTSMOUTH.

376.

PRINCIPLE OF OPERATION

Figure 1(a) shows the basic form of a voiced helium speech wave­form where the amplitude peaks correspond to the s t a r t of the p i tch i n ­t e r va l s . As indicated in the i n t roduc t i on , th i s p i tch in te rva l which is determined by muscular propert ies of the larynx, shows minimal change from normal a i r to high pressure helium/oxygen mixture. Here the helium speech waveform suf fers a s h i f t in (vowel) formant f r e q ­uencies and the rate of decay of the i n t e r - p i t c h waveform is correspond­ing ly more rapid than fo r normal speech. The unscrambler technique considered here consists of s tor ing those sections of the i n t e r - p i t c h waveform Which contain useful in format ion, f igure 1(b) and subsequently time-expanding th i s stored waveform to the general form expected in normal ( a i r ) speech, f igure 1(c ) . Figure 1(c) shows that in the long term there is no time-base change (the p i tch in te rva ls on input and out­put remain equal) however, in the short term, between p i t ch peaks, the time-base is expanded. The discarded sections of input waveform cause l i t t l e degredation to the speech i n t e l l i g i b i l i t y .

The maximum durat ion of the stored segment is governed by the max­imum expected p i tch fo r the input helium speech. The design considered here stores a segment of durat ion 3ms corresponding to a maximum voiced fundamental frequency of 330 Hz.

The degree of i n te r - p i tch time-base expansion is governed by the spec i f i c helium/oxygen mixture being used by the speaker. However, the maximum required time-base expansion is of the order of 3 : 1 . Thus, in order to el iminate loss of any section of input signal which contains meaningful in format ion, four para l le l storage channels are required such that one channel is always avai lab le to store the s igna l . More than one channel may be producing an output at any one t ime, a feature which is acceptable since in normal speech, the sounds produced during succesive p i tch in te rva ls tend to superimpose.

SYSTEM DESIGN

Figure 2 shows a schematic block diagram of the compact helium speech unscrambler considered here. The helium speech input is fed to a high-gain p re-ampl i f ie r which incorporates high frequency pre-emphasis to compensate fo r the rad ia t ion losses produced at the mouth of a speaker in helium. The pre-ampl i f ie r also incorporates an AGC f a c i l i t y (30 db) The pre-ampl i f ie r output enters the analogue delay l ines (Reticon SAD 512), each of which is capable of s tor ing N =256 samples of the i n ­put waveform. As indicated prev ious ly , four storage channels are requi r ­ed to permit a maximum time-expansion of 3:1 without loss of information bearing s igna l .

The pre-ampl i f ie r output fu r ther appears at the p i tch detector . In view of the reduced envelope decay times of the speech waveform in in hel ium, a simple peak detector c i r c u i t can be successful ly employed fo r p i tch synchronisat ion. For unvoiced sounds, characterised by a no ise - l i ke waveform, the p i tch detector operates cont inuously.

The cycle of operation of the process is as fo l lows. The output from the p i tch detector indicates the s t a r t of a p i tch per iod. The

3 7 7 .

input helium speech - of bandwidth up to 16 kHz - ts read into one of the four channels at a clock rate of 85 KHz f o r an in terval of I - 3mS. At the end of th is 3 ms period, the clock frequency f o r th i s channel is reduced by a fac tor which is dependent on the helium/oxygen mixture being used The stored signal is read out at a lower ra te , with an attendent bandwidth compression to the normal 3-4 KHz speech bandwidth. On detection of the s ta r t of the subsequent pi tch i n t e r v a l , the clock and signal multiplexers change over and the next channel reads in the helium speech.

PERFORMANCE

/ Q C l 9 UrSi 3 {?luS5?w s a s e c t l 0 n o f t h e input helium speech waveform (85 ra depth) with f igure 3(b) indicat ing the pi tch detector output def ining a 3 ms storage in terval from the pi tch peak. The loss of signal in the discarded part of each pi tch period causes minimal de­gradation in the output speech. The pi tch period in f iqure 3 i s seen to be of the order of 7 ms.

A *uf1fl?r? ?i a ) s h o w s a n o t h e r section of helium speech waveform (85 m

depth) with the corresponding time expanded output from one of the four channels. Here the i n t e r - pi tch time-base expansion is seen to be of the order of 2 : 1 .

Figure 5 compares the input helium speech waveform (85 m depth), Mgure 5(a) with the corresponding unscrambled output, Figure 5(b). Note that the pi tch in terval (approx 6 ms) remains unchanged although the in t e r pi tch time-base expansion is of the order of 2 : 1 .

Subjective l i s t en ing tests of th i s equipment were performed by a panel of non-expert ( i n both a phonetic and a diving sense) l is teners The test material consists of a section of helium speech recorded at 8b m depth which comprised 15 individual phases spoken once only The i n t e l l i g i b i l i t y to the "raw" helium speech was zero. Figure 6 indicates the measured i n t e l l i g i b i l i t y of the unscrambler output - each l i s tener * w r i t i n g down each phrase as he understood i t . Clearly the mean perform­ance of the l is teners improved as they became accustomed to the un­scrambled output and i n t e l l i g i b i l i t y f igures of 90% are indicated Table 1 compares the system parameters of th is compact analogue un­scrambler with current d i g i t a l real isat ions and emphasises the engineer­ing advantages of fe red by the charge t ransfer device design.

CONCLUSIONS

The helium speech unscramblar design reported here o f f e r s d i s t i n c t engineering and operational advantages in comparison to current ly a v a i l ­able ( d i g i t a l ) real isat ions and the use of analogue charge t ransfer devi technology with associated CMOS logic o f f e r s a design which is suited to rea l i sa t ion as a single integrated c i r c u i t .

ACKNOWLEDGEMENTS

The authors wish to acknowledge helpful contact with Prof H. H o l l i e n ,

c o n t r a ^ U ' S * A ' ™ S W ° r k h 3 S b e e n c a m ' e d o u t u n d e r M 0 D

3 7 8 .

REFERENCES

1 Bei 1 , R.G. 'Frequency analysis of vowel sounds produced in a helium r i ch atmosphere1 J.A.S.A. Vol . 34. 1962 pp 347-349.

2. Holywell ,K. & Harvey, G. 'Helium speech' J.A.S.A. Vol 36. 1964 pp 210-211.

3. HollteruH, et a l . 'Voice fundamental frequency levels of divers in helium oxygen speaking environments' Biomedical Research Col. 4. No 2, 1977 pp 199-207.

4 Giordano, T.A. et al 'Helium speech unscramblers - a c r i t i c a l review of the s t a t e - o f - t h e - a r t 1 Trans. IEEE Vol. AU21. No 5. 1973. pp 436 - 444.

5. G i l l , J.S. et a l , B r i t i s h patent no. 1321313, June 1970.

FEATURE ANALOGUE DIGITAL

Input bandwidth 16 kHz 12 kHz

Power consumption 150 mW 1200 mW

Physical size 10 cb. i n . 1160 cb. i n .

Weight 0.5 lb 17 lb

Standby power 30 hrs 10 hrs

Data frame (T) 3 ms 2.5 ms

Table 1 Comparison of system parameters.

3 7 9 .

PITCH INTERVAL

(a) Helium speech

n 3ms

(b) Stored waveform

(c) Unscrambled output

Figure 1 Pr inc ip le of helium speech unscrambler operation

H E L I U M S P E E C H INPUT

P E A K D t l E C T

. P R E - A M P

CHANNEL S E L E C T

00)

INPUT C L O C K

OUTPUT C L O C K /

C L U C K M U L T I P L E X

CHANNEL 1

CHANNEL I

CHANNEL 3

CHANNEL 4

OUTPUT

Figure 2 System block diagram.

380 .

Figure 3 P i t ch d e t e c t o r o p e r a t i o n . F igure 4 Channel o u t p u t .

(a) Inpu t he l ium speech 500mV/div. (a) I n p u t he l ium speech 500mV/div.

(b) Peak d e t e c t o r ou tpu t 5 V / d i v (b) Channel ou tpu t 500mV/div.

H o r i z o n t a l scale l u s / d i v . H o r i z o n t a l scale 1 / j s /d iv .

(b)

Figure 5 Unscrambler o u t p u t .

(a) I n p u t he l ium speech 500mV/div.

(b ) F ina l ou tpu t 500mV/div.

H o r i z o n t a l sca le 2 u s / d i v .

intelligibility ( / )

Figure 6 I n t e l l i g i b i l i t y t e s t r e s u l t s .

k —

I phrase " 1 3 5 7 9 11 13 15