In continuous speech the acoustic properties (e.g. formant frequenices) of phonemes are affected by the surrounding phonemes and consequently there may be differences as compared to isolated phonemes. I think there is plenty of research reports on this. Start looking articles e.g. from Journal of Phonetics, Phonetica, Journal of Acoustical Society of America, Speech Communication, and IEEE Transactions on Audio, Speech and Language Processing. Try keywords like phonemic variation, allophones, and speaker normalization.

Perhaps I have not expressed myself correctly. Let say a person is saying 'I have a brown dog' and I record the vowel 'o' in dog and study its formant profile. Let us ask this person to keep repeating that statement 'i have a brown dog' up to k-th times. Will the formant profile of 'o' remain the same (let us ignore fatigue). Why am I interested with this question? When we study biological effect of speech sound, we often use recorded voice but this recording is only a single cross sectional sampling. If the formant profile remained the same then I suppose there is no harm to use ths method. I wonder if this has been looked at before and is there a particular term in phonetics describing this phenomena which I could look up.

Now I got your point. Yes, there is some variation, and also publications on this.

We have measured e.g. for the Finnish [o] F1 CV (coefficient of variation) 6.4% and for F2 8,5%, and for the Finnish [o:] F1 CV=8.4% and F2 CV= 9.8%. Values are mean values of 10 speakers, and the spoken words from where the [o] and [o:] were isolated were "toli" and "tooli". The whole article is dowloadable from Linguistica Uralica XLVIII 2012 3.

Eerola & Savela, 2012. Production of Short and Long Finnish ­Vowels with and without Noise Masking, Linguistica Uralica (http://www.kirj.ee/20972/ ).

I think you have two options: 1) use recorded speech, and describe well the analysed sound parameters in the methods part, or 2) use speech synthesis (e.g. Klatt) and describe the synthesis parameters.

This variability can occur and is often referred to as token-to-token variability. This can depend on many factors, such as the vowel sound (some are more variable than others); and the speaker's palate shape (speakers with particular palate shapes may exhibit greater variability). There are a couple of good papers on this, such as:

- http://www.zas.gwz-berlin.de/fileadmin/material/ZASPiL_Volltexte/zp42/zaspil42-fuchs-perrier-brunner.pdf

- http://eresearch.qmu.ac.uk/2450/1/Mooshammer_et_alNij.pdf

Exactly what I was looking for. Thanks

One fact I learned about spontaneous speech is that you may not be able to produce two identical speech tokens in your life. This is so because of the amount of variables involved in speech production, e.g approximations, compensations, aerodynamics, moods, statements, questions, etc. Formant frequencies are the frequency responses of the excited cavities, and this depends on physical size and shape of cavities, the rate of opening and closing, etc. So. if you change the configuration of your cavity no matter how slightly, it will change frequency responses. Looking at this brief presentation of the principle involved, you will have to be a robot or get a robot to speak for you if you want to keep the formant frequencies constant. If you are simply repeating the same sentence time over time, you may have an identical signal (I really doubt it). but if you want to see the effect of cavity changes, place your /o/ between different consonants or vowels and then you'll come to grips with the problem of contextual variability in speech production. Maybe if we knew what you are actually trying to achieve, we could have some ideas to offer.

Akpan Essien (see below) is right, there is not just inter-individual but also intra-individual variation when producing e.g. target words several times. My previous comment (see below) was about inter-individual variation in one study report ( Eerola & Savela, 2012. Production of Short and Long Finnish ­Vowels with and without Noise Masking, Linguistica Uralica (http://www.kirj.ee/20972/ ). I checked from the raw data material of the study the intra-individual variation of Finnish /o/ and /o:/ when producing

the target words /toli/ and /to:li/ five times each. The CV (=SD/mean) of F1 varied between 1.4% - 5.6% and of F2 between 2.6% - 5.2% for /o/, and for /o:/ between 1.8% -3.1% (F1) and 1.7% - 8.2% (F2).

Therefore, if natural speech tokens (e.g. vowels) are used as input in experiments their acoustical properties need to be described in details. What we have done in our experiments, we use over a large population averaged formant and othe values as parameters to produce synthetic speech, and report these parameters.

The previous answers are very appropriate. I'll just add that, in most good research studies, people describe statistical properties of several recordings, not just a single instance. This helps us to step back from the individual utterance, as well as to draw a picture of just the sort of variability you are asking about.

Intra-individual variation during repeating a sentence may have simple reasons: if your mouth gets dry or your emotions change a little bit, the cavities f. e. might be changed in shape, volume a.s.o.

Not only what we call a phoneme, but also what we call a phone is not a physical event but a highly abstract concept.

Akpan wrote; 'Maybe if we knew what you are actually trying to achieve, we could have some ideas to offer.'

My questions was inspired from the following two reports:

Sharma HM, Kauffman EM, & Stephens RE: Effect of different sounds on growth of human cancer cell lines in vitro. Altern Ther Clin Pract 3(4):25-32.

Bailey L: The effects of live versus tape-recorded music on hospitalized cancer patients. Mus Ther 3(17-28), 1983.

When we play taped voice continously, all characteristics in the taped speech sound would remain constant. I suppose. Not sure if someone had actually looked into it. Is that the case in live situation? Bailey showed that even in music it is not the same. I suppose that the premordial sound utilized in Sharma's study would also not be same in live versus recorded setting. In wanted to read further on this subject but could not find the term. Thanks to Sam Kirkhan for pointing that out to me. Now I realize that it is not going to be just formants alone to be considered. There are also other variables to be explore.

The interesting point will be to look whether statistically is it going to be significantly different in live repetition compared to the tape recorded repetition.

Dear Abu Dzarr,

It's true I said if we knew what you were working at, we might be able to help. But it is still not clear to me what your hypothesis is and what you are trying to examine. Your comparison of a continuous tape recorded utterance and spontaneous speech is not appropriate in my opinion. A tape recorded speech, no matter how long it may be, cannot change. Whatever variations had taken place during the speaking and recording session remain constant in the recordings. They cannot and will not change no matter how many times you play the tape. A human speaker is not the same. Every phoneme is perceived categorically because the configuration of speech organs can vary within a wide zone without changing the speech sound. Thus, speaking is not absolute geometry and mathematics. In other words, variability is the rule in speech; but there are very well-known rules that govern the variations, and one that I did not mention before is speech tempo (speaking rate).

Concerning the impact of live music and tape-recorded music on patients, I do not seem to see why you pick on formant frequencies. There are loads of parameters in a piece of music that a patient might react to, e.g. pitch modulations, tempo, intensity, sound quality, frequency range, etc. Another factor is the acoustic instrument (strings, wind, percussion, etc). Vocal music therapy is more than formant frequency. I;m still trying to understand your project.

Akpan.

Vowel formant frequencies depend on preceding and following consonant of the syllable. Generally these variations are presented with scattering ellipsis in the F1-F2 and F2-F3 plane. You can read Acoustics of children’s speech: Developmental changesof temporal and spectral parameters

Sungbok Lee,b) Alexandros Potamianos, and Shrikanth Narayanan