Pitch Analysis - A fun tool to investigate Shruti in CM

cmlover · Post by **cmlover** » 20 Dec 2007, 07:54

We have been having a number of discussions on shruti. There have not been many objective analysis on it barring some pilot studies by Arvindh. It is to be admitted that he human ear is notoriously misleading as we know from a number of experiments in psycho-acoustics. I have tried to adopt a clinical tool which is used by clinicians in speech analysis to analyze svara components of CM and presenting some conclusions. If there is any interest I can do more analyses (time permitting) which may be a 'myth-buster' and may even be controversial. I shall not however reveal the names so that the privacy of the artistes are never violated. Also I admit at the outset my inadequacy of 'theory' (being a 'dabbler' in trying to interpret technical areas relatinv to CM.

The following is the svara pattern of the raga 'bahudari' of a famous female artiste (she is singing at 5 kaTTai (G) as you can see). Tha aalaapanai was for about 8 minutes and hence there is adequate sampling of the svaras. Here is the pattern.

The same artiste has rendered 'Brovabharama' following the aalaapana. And here is the svara pattern of the kriti rendering

Bahudari is a janya of the 28th mELam with
aro/avaro as S G3 M1 P D2 N2 S' / S N2 P M1 G3 S

The riShabha varjyam is quite evident in both plots. The aalaapanai is quite consistent with the svaras though there appears to be a slight excess use of D1. It is quite likely that during the kambitam of panchamam and dhaivatam the artiste strays into the anya svara. However it is difficult to detect the excursion during the aalaapana which was quite smooth and commendable. However a 'picky' reviewer may be able to detect the excursions and fault the artiste.
The kriti svarams are much worse where the N2 is competing with D1. Is it acceptable? I do not know. However if one were to focus on the sahityam which was rendered 'impeccable' one will not notice the lapses. In amny cases artistes do use the sahityam as a 'shield' to escape from shruti lapses. But critics do notice the shruti and they are trained to ignore the mesmerising influence of the 'words'

Your comments.....

arunk · Post by **arunk** » 21 Dec 2007, 16:51

cml - a few points

Even with the cloud of anonymity, such analysis' make me quite uncomfortable. The only solution is to analyze something, for which the artist/performer is explicitly ok with. Maybe you can ask one of the rasikas here to volunteer

. Either their own, or you could ask them to "mimic", a small portion of a professional recording. I think if the piece is not elaborate and complex (which it mostly wont if it is short), many amateurs can do a very good job.

I know this is a bit dicey but I really think that is the only safe and "nice" way.

Having said that,
* I do not know how to interpret your pictures/images.
* We need a pitch curve not discrete slots (although at the bottom most level of computer based pitch estimation, it is indeed discrete slots - but that is at a 20-50 ms level). This is very crucial particularly for CM where melodic continuity is paramount and pitch transitions are very high - both across swaras, as well as within swaras (gamakas). In essence, without a curve, the conclusions "excessive use of D1" appear without proper support. We need to know what are the time-slots that show this, and whether
a) the actual signal is clean as in without disturbance from accompaniments
b) is it in the middle of a gamaka
b) does it involves attack phases, or places where a consonant starts. Both can throw pitch estimator quite a bit.

Arun

cmlover · Post by **cmlover** » 21 Dec 2007, 20:26

Thanks Arun
You raise a number of valid points. Pitch analysis in CM is highly complex. Having said that the same problems persist even if an analysis is reported by a 'critic'. In most cases 'reviewers' criticize performers for lapses without citing the context. Again we automatically assume that most rules regarding 'svara patterns' in ragas are immutable and sacrosanct. While humans suffer from 'psycho acoustics' computers mercilessly will detect deviations albeit out of context at times. I always wonder about the lattitude of swings during gamaks which is never rigorously quantified. When is it a 'beauty' wnd when does it get 'ugly' is highly subjective.

It is nice to get a peek at you in our rasika crowd and I felt being there with you all. Have a good time and convey my greetings for the newyear to one and all. We will continue our fascinating experiments after your return when I hope Suji also will be back. Cheers.

Anonymity is needed for an objective analysis and to break myths perpetuated in CM by self-styled experts. Let us hope CM from being fine arts will mature into a fine science in this century with the contributions of enquiring 'scientist' like you, Uday etc.,

Have a great time!

vijay · Post by **vijay** » 04 Jan 2008, 14:01

CML just saw this. Share some of Arun's concerns but very interesting nevertheless. Maybe we can discuss how we can refine this so that it an help the art and also understand it better...Arun's curves seem to make a lot of sense...can we amp out phrases and gamakas in varioous ragas using a software...

I dread to think of the day when a swara decoder is placed in front of artistes and displayed on an LCD panel for all to see with a wimbledon-like "beep" whenever the artiste crosses a threshold!

balsree · Post by **balsree** » 04 Jan 2008, 15:25

vijay wrote:I dread to think of the day when a swara decoder is placed in front of artistes and displayed on an LCD panel for all to see with a wimbledon-like "beep" whenever the artiste crosses a threshold!

And a tala track-o-meter which automatically detects what tala is being used and buzz on lapse.

Ideally, I would really like to see a mechanism which will not let a rasika get up in the middle of a concert esp during thani. May be a magnet that will attach them to their seats till the end of the concert !!

rajumds · Post by **rajumds** » 04 Jan 2008, 20:17

We may be listening to beep & buzz cutchery only many a time

vasanthakokilam · Post by **vasanthakokilam** » 04 Jan 2008, 22:59

cmlover · Post by **cmlover** » 05 Jan 2008, 02:55

My goal is to look at CM objectively. Of course an art form is for enjoyment and cannot be dissected and crutinized using the surgical knife of science. But then when myths are built around personages and self-styled critics shoot down up-and-coming artistes using illusory weapons one needs to rebut and set right the rules of the art-form using science as the tool. My admiration in that context was for Arun who started out as a critical analyst but now sadly is backing off quoting abstract verses from ancient texts and abandoning the very same scientific tools that he himself forged. Let me get down to some facts.

Many times 'famous artistes' during manodharma reel out svaras which will not stand a cold critical shruti analysis. For example he/she may sing G M P D N S' but a shruti analysis will show that it is actually G G P G R S. But our ears tuned to the 'name' of the note will not notice the lapse. This happens more often in madhyama and durita kaalam. The errors do occur among instrumentalists too but are less common. Should we say that it is Ok since 'to err is human'. On the other hand even if it is shrutiwise G G P G R S while singing Mohanam he said G M P D N S' there will not be a single critic who will not pounce on him.

My question is should we be sicklers about shruti ( also laya which is the other independant issue!) ?

ramakriya · Post by **ramakriya** » 05 Jan 2008, 04:02

cmlover wrote:For example he/she may sing G M P D N S' but a shruti analysis will show that it is actually G G P G R S. But our ears tuned to the 'name' of the note will not notice the lapse. This happens more often in madhyama and durita kaalam. The errors do occur among instrumentalists too but are less common. ?

cmlover,

The chances of the above exact situation happening, and ears not detecting it is as slim as finding drinking water in the sea

I might have agreed if u said something like G M P G R S is utterd like G G P G R S and vice versa - but definitely not the specific example you cite!

-Ramakriya

coolkarni · Post by **coolkarni** » 05 Jan 2008, 04:57

see with a wimbledon-like "beep" whenever the artiste crosses a threshold!.

Vijay..
And then you could have the Questioning calls too...Imagine the artist standing up , walking from one corner of the dias to another,waving to the crowd,sometimes imploring head in hand, yelling like Mcnroe...
and then softening down and continuing with his rendering of the krithi.....

Nindakaru Irabeku......

cmlover · Post by **cmlover** » 05 Jan 2008, 07:15

Ramakriya
In many countries now-a-days drinking water is found in sewage water:
Better to have sea water

I may be able to locate what exactly I said but it will take time because of the mess in my filing system

Now look at the following mohanam kalpanasvaram by one who was most famous for the same. (you can try to guess but no names please.
http://www.mediafire.com/?0frnytmthkl

Here is the plot as analyzed objectively by the speech software:

Obviously he (yes it is he since I have ascertained his shruti at 2.5 kaTTai!) is singing
PGRSRG,R,SGRS
Making all allowances for gamakas what is the mantra nishaadam and dhaivatam doing here? Could you recognize mohanam from the plot, though the audio will fool you into guessing mohanam. If not the artiste will bully you like McEnroe to accept it since his reputation was made on it

What am I missing here?

vasanthakokilam · Post by **vasanthakokilam** » 05 Jan 2008, 10:17

CML, something is wrong with the analysis software, I hate to say. It shows as if he is singing something that is Vivadhi raised to the power Vivadhi

. It shows almost all notes of the octave. Do those percentages mean anything?

Also, in my limited experience with looking at anything based Fourier series analysis, there are all sorts of frequencies that show up,especially the higher harmonics. You have to filter out a lot to get a realistic picture that is consistent with what one hears.

For the same clip, can you show the view with time as X axis and Y the frequency. I am referring to the picture which shows a stacke pattern of colors including all the higher level harmonics. My guess is that, if you look at the picture, it will be much much closer to the swaras the singer is spelling out.

cmlover · Post by **cmlover** » 05 Jan 2008, 11:13

VK
This is a software used by clinicians and I am not sure about the basis and can only guess (nor do I have the source codes!). I believe it calculates the transit time in the frequency ranges of the fundamental (F0) as below
131Hz - 138Hz is 'D2/'N1
139Hz - 146Hz is 'D3/'N2
147Hz - 155Hz is 'N3
156Hz - 164Hz is S
165Hz - 174Hz is R1
175Hz - 184Hz is R2/G1
185Hz - 195Hz is R3/G2
196Hz - 207Hz is G3
208Hz - 219Hz is M1
220Hz - 232Hz is M2
233Hz - 246Hz is P

since there are 13 notes here roughly each note should get 1/13 or 7.7%
If you take P for example, the histogram shows roughly 5% in P and the remaining 2.7% 'smeared' onto M2 and elsewhere. Again R2 should receive 5/13 or 38% whereas the histogram shows about 23% with the remainder 'smeared' on to neighbouring swaras. That is quite natural since there is a transition time from one note to the other. But then If the shruti control is strict then the smearing will be minimal. I have tried this software to try to detect ragas (where the aro and avaro are same notes) and it performs quite well for 'some' not so well-known performers who have excellent shruti control. I can easily detect melakartha using this software. Also instrumentalists do quite well especially flautists

vasanthakokilam · Post by **vasanthakokilam** » 05 Jan 2008, 12:59

CML, thanks for the background info.

With that kind of 'smearing', how can you even ponder questions like 'what do mantra nishaadam and dhaivatam doing here?' Both 'Ga's got close to 20%. There is a small Ri which has higher % than Sa. Is that smearing or real? Same question about both Ga s.

Also, you seem to be completely discounting what you hear and go by this 'objective' software when what it depicts is vulnerable to all this smearing and trust what it shows is real swaras.

cmlover · Post by **cmlover** » 05 Jan 2008, 22:02

Agreed! That is psycho-acoustics!
You hear what your mind wants to hear; not what the ear hears!

vasanthakokilam · Post by **vasanthakokilam** » 05 Jan 2008, 22:31

You hear what your mind wants to hear; not what the ear hears!

CML, I just do not buy it.

I stil do not understand why you are putting such 100% faith in what the tool shows. I am limiting this statement to just this instance.

You have not addressed my questions about smearing. Meaning, in this instance, what made you interpret that there is mandhra NI when you did not make the same claim about all the other 11 different swaras that the tool shows.

cmlover · Post by **cmlover** » 05 Jan 2008, 23:14

Oh sure! I am worried about the other extraneous svaras too! But they are in-between and one has to traverse through them. But the mantra svaras are outside the 'legal' range and to invoke them is indeed a shruti lapse!

vasanthakokilam · Post by **vasanthakokilam** » 06 Jan 2008, 00:17

CML: With the limited data you have shown us, all I can say is that your faith on the output graph in this instance is an order of magnitude more generous than what it warrants. Of course you know more about this than me, so I will leave it at that. I think you will agree that knowing the limitations of the tool for a given application is the most significant piece of any analysis and I am not convinced we know the true happenings behind the tool.

If you can show the frequency spectrum graph ( the other tool you used to use ) for the same clip, may be that will help, otherwise this is a dead end.

cmlover · Post by **cmlover** » 06 Jan 2008, 01:44

PGRSRG,R,SGRS

Here above is the FFT plot. The notes show the gamaka patterns. The first P has been swung too much to have smeared M2 and definitely M1 from where G1 takes off. The gamakas on R and G are moderate and acceptable. The S (10th note) has been wildly swung which resulted in the mantra nishadam and dhaivatam being touched. Somewhat similar (but less moderate swings in the final S as also the 11th G could pass for a S. (Note that the bottom blips under the last 4 notes are due to the drum excitement).

Are there 'legal' gamakas for the shadjam in mohanam? Can we justify shruti lapses claiming that they are due to gamaka swings?

By the by I am not mesmerized by the tool which is a useful valid clinical diagnostic tool, but am trying to be objective.

vasanthakokilam · Post by **vasanthakokilam** » 06 Jan 2008, 06:16

CML, This is much better. The singer seems to sing PGRSRG,R,DSRG and the plot more or less matches that ( though it is really hard to tell with other frequencies overlapped ). Also, the mridangam frequencies are overlapped which makes it hard to separate the voice frequency. But the overall pattern of up and down seems to match and that is the extent you can read from the graph.

If you can, please do the following.

a) Include the Y axis frequencies.
b) Adjus the Y axis so that it only shows the main octaves and not the higher level harmonics. This will blow up the region of interest considerably.
c) Increase the sampling rate to a high enough level with a known set of tones for calibration purposes and then do the graph for this bit.

The spectragram at this site is very good and I have had good results ( there is a 10 day free trial )

http://www.visualizationsoftware.com/gram/gramdl.html

cmlover · Post by **cmlover** » 06 Jan 2008, 07:18

Here it is. Remember there is a limit to the resolution due to the sampling rates

cmlover · Post by **cmlover** » 06 Jan 2008, 08:04

Now here is the histogram of just PGRSR

And here is the FFT calculation as I could extract from the system

Just note that the red lines are the F0 approximations which are counted into the histogram bins. Clearly the frequencies are smeared all over and I am sure it is not due to 'drum' artefacts. What do you think?

vasanthakokilam · Post by **vasanthakokilam** » 06 Jan 2008, 08:45

I have just calibrated the frequencies upto 640 at 100 Hz intervals for ease of reading. That is all we need. If you start with PA at around 456 ( or a gamaka which includes Dha at 512 ) and trace the contour you pretty much track the following frequencies as the graph meanders through.

P - 466 ( really P G P )
G - 392 ( slide from P to G and then steady G )
R - 350 ( sort of P R )
S - 311 ( steady S )
R - 350 ( bit of a wavy R )
G,-392 ( wavy G setting up for the next maneuver )
R,-350 ( through a complicated gamakam that takes it all the way to D and settling in R. It looks like G D R G R. sounds nice in the vocal )
D - 524 ( It is really G P D P settling down on a neat and steady P at 456 )
S - 311 ( it is a complicated move from D to S through P and intermediate notes )
R - 350 ( there is a bit of S R S there )
G - 392 ( nicely shown as the wavy line just below 400 towards the right most end of the graph with a little blip to P and then G )

There are so many lines there, so use the above numbers as an approximation to track through the graph. I say the singer is pretty close given the mandatory Mohana Gamakam and the mridangam sounds.

If the graph shows up as too small, save it and blow it up a bit with some kind of JPG viewer.

( If you can redo the graph maxing the frequency upto 640 and draw horizontal lines at every 50 Hz, then it can be much clearer. )

vasanthakokilam · Post by **vasanthakokilam** » 06 Jan 2008, 09:20

Same picture as above but with the focus on upto 640 HZ and with the contour of the melody marked in red. CML if you can redo the graph like this with your program, it will be much clearer. This has a ghostly quality to it since I edited it out with Paint.

cmlover · Post by **cmlover** » 06 Jan 2008, 11:34

VK
Here is the calibrated plot

But note that F0 is half the frequency which is not seen in the plots. I have given you the plot of the F0 (shruti) calculations. We are looking at the transit times in the specified frequency ranges and not whether he hits the right frequency . He swings wildly around the stipulated range due to gamaka fluctuatons (aandholika?). The HM folks criticize CM stating that the performers hide their unsteady shruti using the 'figleaf' called gamaka

vasanthakokilam · Post by **vasanthakokilam** » 06 Jan 2008, 11:57

CML, please adjust the Y axis, so that it goes only upto 640Hz.
And create a grid with a horizontal line at every 50hz. That will make things very clear in that blown up view at good resolution.

I am going by your original statement that the singer sings at 2.5 kattai. My frequency numbers are based on that. I am also assuming that the frequency plot is as capture and not adjusted for any sruthi by you. If both of these are not correct, please alert me to that.
(I am confused about F0 etc. but that can wait.).

And at this point of the analysis, we can not afford to care about what a generic HM artist calls a generic CM artist. That is not material now. We can get back to that later if this analysis provides a reasonable basis to refute/corroborate such things.

cmlover · Post by **cmlover** » 06 Jan 2008, 20:53

Done. Ghosts abound due to sampling...

What is your point?
(for the shruti of 2.5 the aathaara shadjam for the range under consideration which is F1 here is
312Hz to 329Hz. You can calculate the frequencies for the other notes!
Here is the Table for convenience
262Hz - 277Hz is 'D2/'N1
278Hz - 293Hz is 'D3/'N2
294Hz - 311Hz is 'N3
312Hz - 329Hz is S
330Hz - 349Hz is R1
350Hz - 370Hz is R2/G1
371Hz - 392Hz is R3/G2
393Hz - 415Hz is G3
416Hz - 440Hz is M1
441Hz - 466Hz is M2
467Hz - 494Hz is P

vageyakara · Post by **vageyakara** » 06 Jan 2008, 21:53

CML gurujiIt shows to prove how much paistaking effort you have put into this subject
Ramaraj

vasanthakokilam · Post by **vasanthakokilam** » 06 Jan 2008, 23:51

CML, Thanks for adjusting the Y axis scale and calibrating it at 50 hz. It is much more visualizable now.

The frequencies I used for the swaras and the one you provided are close ( My first attempt last night was off a few Hz which I have fixed now ).

You have provided a range of frequencies for each swara. What does that signify and how did you arrive at that? ( First I thought you are putting in the frequency for both equal temperament and just intonation but that did not quite match ).

My confusion about F0 and F1 is that for 2.5 kattai, Aadhara Shadja is at D#. In fact the frequencies you have arrived at are based on D# as aadhara shadja. (isn't it?, If not please set me right ).

---------------

Back to the clip, my original write up about the meanderings of the main melody line is still good and I call it Good Mohanam with all the gamakas with out any sruthi slip. Do you still contend that there is Sruthi slip in this bit?

To me this exercise is more on how to read and interpret the frequency graph for CM (with all the gamakams and smooth transitions ). Once we get a baseline on that, we can then use that to compare other snippets of music.

cmlover · Post by **cmlover** » 07 Jan 2008, 01:12

Off to the Temple! Will answer you tonight!

vageyakara · Post by **vageyakara** » 07 Jan 2008, 04:04

VK ji, MOHANAMMWITH GAMAKAS !! I could nt relish that idea

cmlover · Post by **cmlover** » 07 Jan 2008, 05:34

VK
Let me explain as I understand. The basic calculations are based on the equitemper scale on which the Piano is based and which is used for KaTTai(shruti) calculations in CM. The Octave used for this calculation is from 131Hz to 261Hz which is divided into 12 semitones. The fundamental (F0) calculated within this range determines the shruti. For example 131Hz 138Hz will be one KaTTai shruti, 139Hz to 146Hz will be 1.5 kaTTai shruti and so on. This artiste has his fundamental (F0) in the range 156Hz to 164Hz whence he is singing at 2.5 kaTTai which is his aadhaara shadjam. His taara shadjam will lie in the rnge 312Hz to 329Hz. The FFT graph shows clearly only the second octave (which is F1 and hence we can decide on the svara sung from the graph using the table I provided in #27.

Now accordingly when he sings P his voice range should be 467Hz to 494Hz. The FFT graph tells you that the P ranges between 350Hz to 500Hz. Though this included the correct P he swings all the way up to R2. Question is, can this be explained as a gamakam swing or is it unsteady shruti? From the wave plot (seen at the top of the graph) we know that his P lasts for about 400ms (milli seconds). His next svara is G3, and if he is getting ready for that he should not swing below 393 Hz. My clinical speech software can dissect and analyze the components of this P (~400ms) and create a histogram which is shown below.

You will notice that he is spending more transit time in M2 and G2 than in P itself. This is what I termed as 'smearing' the svara. We can also analyze the other svaras some of which will be right and tight. But there will not be a consistency. As far as I know there is no gamaka on P in Mohanam (which our Ramaraj can confirm).

Our human ears are not tuned to detect these lapses and partly due to the 'psycho-acoustics' (i.e we hear P when he says P though it may not be true

Now you try and explain.....

vageyakara · Post by **vageyakara** » 07 Jan 2008, 07:46

Panchamam is an "ACHALA SWARAM" in any ragam just like Shadjam.If at all , it is deliberately oscillated it is attributed to cheap gimmidcs and nothing else.

vageyakara · Post by **vageyakara** » 07 Jan 2008, 07:55

iN THIS CONNECTION I would like to quote SRJ -(in tamil) thus:-
"ella swarangaLayum poTTu asachundE irundA kaTTaDam ATTam kanDu viDum"
meaning that the entire structure will give way , if the pillars are constantly made to shake
Here the pillar is referred to Sapthaswaras and the structure as musical creations.

vasanthakokilam · Post by **vasanthakokilam** » 07 Jan 2008, 09:01

CML,

First a quick note on the calibration, you and I are on the same page. Your frequencies are absolutely right. I am still confused why you call 131Hz as F. Because in Equal temperament Middle A is 440 Hz. By this standard, 131 will be C. May be when you say F, you just mean to say 'F'undamental and not the note 'F". Anyway, it absolutely does not matter at all for this discussion since we are on the same page with respect to what frequencies we are calling S, R G P D S.

Now to the analysis.

1) I am quite wary about making pronouncements on the singer. Hope you understand my reluctance to get there quickly. I want to get a firm footing on reading the ghostly image of the frequency graph. Even a steady S is not a thin line representing one Hz. I know you are being quite generous in giving the whole interval to a swara. This is just a general remark on the iniital purpose of the exercise. I see it more as calibrating the graph and a general methodology of reading off data from the graph.

2) Let us take your first assessment about the P as to why there is so much smear and swing Let us just focus on that for a second and nothing else. The singer starts at P. We both agree. I also agree with what you read off the graph subsequently. Next, If you listen very carefully, at the 100 msec interval from the beginning there is a bit of a mettalic sound that does not belong to the singer. Fortunately, there is no Mridangam sound when the singer sings 'P'. Mridangam sound starts at 'G'. I contend that either the Singer is doing a nokku type thing and actually sings P G P. And the smear down to R2 is contributed by the mettalic sound at the beginning. This is just a guess. Or may be the singer is really intoning his P to be a smooth P R G P but if that is really the case we will hear it. May be some one else with a much better sense of frequency in short timespans can comment.

3) Having said this, analyzing polyphonic sound with FFT is quite tricky. There is very limited polyphony here so that is better ( Singer's voice, Mridangam, tambura to deal with ). You have picked a good piece to analyze since there is no violin. But we have to account for the Mridangam and tambura sounds as we move to the right of the graph since there is much much more smearing. Let us not quickly jump on the singer. That is the last resort after we eliminated all other possibilities.

4) Finally, it is indeed intersting to analyze ( hopefully with other samples ) the correspondence between the Solfa letter the singer sings and the frequency distributions, oscillations and glides. I am not talking about Sruthi slip here but the very essence of gamakams.

CML, in this context, I would like you analyze and put up a picture of the Mohanam Gamakam on G that MSS sings as part of the chittaswaram in Bhavayami. It is at the very beginning of the swaram and it is quite unique. I would like to see what countours it takes. I am more interested in knowing what frequency contours make that excellent sounding G gamaka ( and not whether it is part of Mohanam or a sruthi slip etc. )

arunk · Post by **arunk** » 07 Jan 2008, 09:09

Some points
* From what I can read, sruthi (sa in madya stayi) maybe around 160 Hz - not 320, Look at the SPACING of the harmonics around 1 second (i.e. after the + mark). You see 320, 480, 640. Even if the 160 component does not exist, our ears will perceive the pitch as 160 - the spacing is all that matters. Hence pa (madya stayi) is approx 160 * 3/2 = 240. A steady flat Pa should include harmonics of 240, 480, 720, 960 etc. But 160 is high for a male (at least this male

).
* Note that for ga - should be 160 * 5 /4 = 200 - there is strong 160Hz component. IMO, that is because of the mridangam component (tuned to sa). This will also result in 320, 480 etc. show up. In general you have limited polyphony here
* The pa here is delivered with an emphasis (hence delivered from above - p(d)p or p(Sd)p), and that itself results in a pitch curve - expecting a flat pa here is an (edited) incorrect expectation. It is a mistake to assume that since sa and pa are constant, they will always be delivered flat AND DISCRETE (like WM) from previous/succesding notes, You can slide to sa and pa, you can deliver pa with emphasis like here. Note that short-time FFT assumes that for each window the signal is harmonic and that exact signal extends from -infinity to +infinity in time. Thus a signal whose pitch varies in time is not exactly the ideal signal. The way to compensate is to have sliding overlapping windows - which I would assume is used here (this will allow you to see a pitch curve but one has to be careful to know its limitations - see next point)
* Like I mentioned when swaras like pa is rendered, a small but significant (for short ones) portion of the first part is unvoiced due to the consonant "p". Since that part is not harmonic, considerable smearing WILL BE shown by FFT. Note that for unvoiced and unharmonic portions FFT is trying to force-fit a harmonic model when it does not exist in the signal itself. Such portions can be safely discarded as it is pure junk - the model works only for harmonic parts (like the steady portions like you find in the akara part).
* It is also a mistake to assume swaras (even during kalpanaswaras) are discrete, flat entities most of the time (i.e. swaras are not notes) In fact most frequently it is not so.

It is also obvious that in general the signal is not perfect for the analyzer has significant smearing is shown even for steady portion of the elongated ga note. The ri is slightly better.

Arun

arunk · Post by **arunk** » 07 Jan 2008, 09:10

I cannot tell which line is which Hz (very poor at reading but the axis markers can be a whole lot better) and so my 160 may be wrong

cmlover · Post by **cmlover** » 07 Jan 2008, 09:23

I am sorry for the confusion. F standa for the Fundamental and not the svara F.
I agree one has to be cautious about these critical analyses. My biggest concern is the sampling fluctuations. Just note that Arvindh has never discussed it in his papers when he talks about the subtle svaras. FFT is a powerful tool and we should use it to analyze CM svaras. I am not putting down anybody but only drawing attention to the theoretical aspects of CM which have to keep pace with the scientific development and tools of the 21st century. CM is a beautiful mathematical concept in addition to being a fine artform. The theories should be mathematically cast and whetted using latest computer tools rather than depending on and quoting the 'sutras' from old abstruse texts. Take your time. You spout bright ideas!

vageyakara · Post by **vageyakara** » 07 Jan 2008, 11:11

Dear CML guruji,
though I am not competent enough to even go nearby your eloborate technical studies , I wish to add the following lines.All your illustrations, and detailed analysis are by default, leading towards one supreme universal truth that governs the "SHRUTHI ANANTHAM theory.I think we have discussed this during your visit to Chennai some months back.
Even a fall of an object(let it be metal, glass, wooden pieceand a cloth bundle etc.,) would produce a sound (dhwadni) which may or may not get fixed into the presently available SHRUTHI scheme.
When likewise , Adhaara shruthi itself is subjected toi variations, there are unfathomable scopes for eloboration and elongation.
What is available with us today (22 shiuthies) is a regularised system from where the mEla schemes emerged.
Due to evolutionary changes , which are applicable to any form of fine art, our system of CM is no exception.
And now coming to the frequencies part of shruti aspect, it is bound to oscillate, and how much , is a pertinent question. It depends on the qualititative capacity of human hearing device (EARS) to be the best judges.
Perhaps many of us might have come across the following innovations tried by our ebullient vidwans and vidhushis:
( 1)Producing Sapthaswara sounds on the Mridangam, tabla and other percussion instruments.
(2)Bringing gamaka effect in JALATARANGAM by just stirring the water in the bowl.
(3)Flutists giving a differential effect by just keeping a slanting position and blowing different levels of wind through their flutes.
(4)Violins adopting diferent techniques to produce melody by different level of presure applying to their fingering subtlities.
It has thus become evident that different levels of intelligentia have resulted in different levels of performing prowess.
And finally hats off to u Sir for the meticulous endeavour you put in to specify the frequencies in numeral terms.And no effort will become a Waste in so far as CM is concerned
regds.
Ramaraj
t

cmlover · Post by **cmlover** » 07 Jan 2008, 19:53

Thanks my dear Ramraj
I distinctly remember and enjoy our discussions. Shruti (and of course laya) are fundamental to CM. We need precise definitions and "objective" tools for monitoring them and human ear alone is a very imprecise mechanism. For example it is pseudo to claim that the Adhaara Shadjam(AS) is immutable during a concert. Most often as the artiste gets tired his AS varies which he will try to adjust, and if unnoticed the accompnists slyly retune their instruments to correct for the variations. The Rasikas too get tired over time and their perception will vary over time. Voice training is of paramount importance in HM. For WM the precision required is astounding and there are professsionals whose only job is to tune the instruments (eg Piano Tuners). In CM we mostly depend on the ears and take for granted that it will be OK and our tolerance is very high except when a 'critic' chooses to pounce on an aging artiste or an up-and-coming youngster for shruti slippage

On second thought, it is not a bad idea to build into those Radels (which are quite common now on stages)a feedback mechanism to alert the performer when he/she slips on shruti (and laya too!).

vgvindan · Post by **vgvindan** » 07 Jan 2008, 20:04

The theories should be mathematically cast and whetted using latest computer tools rather than depending on and quoting the 'sutras' from old abstruse texts.

I am reminded of the surgeon who went about with knife and scalpel to find the soul.

vageyakara · Post by **vageyakara** » 07 Jan 2008, 20:12

Excellent idea CML Guruji,
i.If a device as suggested by u is discovered it will be a boon to the vocalists.Besides ,they should pay some attention to voice culture as is done by their North Indian counterparts.And finally any artist who wishes to pursue CM or any Music seriously as a profession should , besides leading a pious and disciplined life which is althemore important to present an emulative personality.
Even during olden days many of our JAMBHAVANS lacked the all important factor "SHRUTHI SUDHAM" miserably.Only SAAHITYA BHAVAM on many an occasion came to their rescue.Some took to LAYAM expertise and some were proficient in Raga alaapanai,swara swinging etc.All in one combined vidwans were very hard to find.
At least during their Sadhakams, a device (as suggested by u) can be tried by music aspirants and attain perfection gradually.I am putting forth the above lines jusat as suggestions and nothing else.

cmlover · Post by **cmlover** » 07 Jan 2008, 20:42

Arun
I buy your comment on the effect of the consonant which can distort the shruti. Perhaps svaraprstaarams should be done purely on akaaras rather than using solfas (which will be a major break in CM

But truly the real melody is in the instruments and human voice with all its variations has an indifferent quality. But may I ask whether the consonants can be used as an excuse for shruti slips?

VK

will give it a try. First I must get MS's shruti....

cmlover · Post by **cmlover** » 07 Jan 2008, 20:45

VG
If the sugeon is careless about his knife or scalpel you lose your soul!
Just a reminder

cmlover · Post by **cmlover** » 08 Jan 2008, 08:13

VK
It looks just terrible! Should I post the results since some 'fanatics' may get offended!

arunk · Post by **arunk** » 08 Jan 2008, 08:27

cmlover,

IMO, there is perhaps a couple of fundamental problems with some basic assumptions you are making that is leading you to premature, incorrect conclusions (or raise questions implying such conclusions):
1. You are assuming that all frequency components you are seeing is exactly there in the vocal part of the signal (the only part you are interested in) when in truth there are many factors that can prove this otherwise: Polyphony - as in mridangam, tampura, violin; Noise (even if our ears can filter this out, they can be at a level that can show up significantly on spectrograms); Limitations of FFT itself
2. To know something is "off shruthi" one must understand how the gamakas appear as pitch curves in FFT. You started off with the assumption that FFT can show minor shruthi slips that the ears can miss, and more importantly any variation that "doesnt look right" is really there and is a miss. Now, FFT will show minor pitch variations that most (if not all) human ears can miss, because of #1 above it can show wild variations which most musically trained ears will catch. Before one can label what is a shruthi slip or not, one must be very clear on two things:
a) what is the pitch curve expected? Note that there is no single exact curve for a gamaka (arvindh shows this and it is definitely true). So what the expected set of curves (or general shapes).
b) Is the FFT result reliable for that particular signal at that particular spot? Does it have noise components? Was it thrown off by polyphony, consonant starts and attack phases?
Of course there can be variations which are beyond the threshold of normal/average human hearing - i.e. natural tremolo of the human vocal chord. Note also that for male voices in mandra stayi if the voice goes a bit "kara-kara", the harmonic nature comes down a lot, and once again FFT can get quite confused and show all sorts of junk. The reason you would find flute better is because it is higher pitched and a very harmonic sound. Female voices can be better. Male voices are hit and miss. This is just from my experience.

In any case, without knowing (a) and (b), it is not prudent to jump or even ask questions about "is it a shruthi slip". You/We should first ask
1. Did it sound proper to our ears?
2. Are we sure that the FFT results reliable for this particular signal at that particular time? (how much smearing is there? Is there interference from noise and other things? Are those interferences separatable?
3. What is the pitch curve - expected vs.shown?

I also think by the kind of conclusions you are leaning towards, vgv's comment makes a lot of sense. While I know where you are coming from (difference of opinions about shruthi lapses etc.), the kind of questions based on the FFT analysis of this signal alone (which I think is poor - i.e. signal not ideally dissected), vgv's counterpoint rings quite strongly true

So I suggest a different and more cautious approach.

Arun

cmlover · Post by **cmlover** » 08 Jan 2008, 09:23

Thanks Arun for that clear conscious analysis. My analysis are primitive and I am the first to admit that. But then we need to forge tools to study the problem or find explanations to obeerved phenomena. I do not agree that VGV's philosophical approach will help science or CM either. Just claiming that 'God knows and let us leave it to him to do his best ' is quite irrational and a fatalistic approach.
You have raised very valid points. Polyphony is a vexing problem in piitch analysis. But claiming that 'it is possible' does not mean that that is the problem. First note that the drum is apparently not there since its signature appears only later. The violin also is inaudible. The dB cut offf appears to be around -42 and if they are below this level they will not be included in the calculations. Computer takes into account only what is physically there withinlimits of detection whereas the human mind can create what is not already there. I will prefer an objective approach to a subjective perception analysis. Your explanation of physiological factors are important vis-a-vis instruments and that is precisely the reson for the slip ups.
Finally I am approaching the problem with an unbiassed mind and not to put down or villify any artiste. If the analytical approach (not necessarily this one) can stimulate thinking and make CM bettter and more enjoyable my mission is fulfilled.

vasanthakokilam · Post by **vasanthakokilam** » 08 Jan 2008, 14:18

cmlover wrote:VK
It looks just terrible! Should I post the results since some 'fanatics' may get offended!

I assume you are referring to the MSS G in Mohanam in bhavayami. Yes, I would like to see it, We all know for a fact that it sounds nice, great and correct. It is more about how FFT shows a good G rich in Gamakams. ( so no conclusions, even tentative, on the artist based on the FFT.

)

vasanthakokilam · Post by **vasanthakokilam** » 08 Jan 2008, 14:25

But may I ask whether the consonants can be used as an excuse for shruti slips?

I think what Arun said is, with the beginning part of a consonant the harmonics are not rich and so FFT is not going to be very useful. That observation only applies to the non-voiced portions of a consonant and not the full consonant since the ending portion of the consonant has a vowel sound, as in solfa like S, R , G etc. So it should not have anything to do with any sruthi slips.

vgvindan · Post by **vgvindan** » 08 Jan 2008, 17:32

I do not agree that VGV's philosophical approach will help science or CM either.

cml,
I did not want to enter into this technical discussion. However your rubbishing Indian Sutra system -

'sutras' from old abstruse texts

- is very distressing from a person who should know better.
While any fanatic approach to Indian philosophical system should be shunned, it is also true that a whole-sale dismissal of our philosophical thought system is not conducive to Scientific enquiry. A true scientific enquiry will not dismiss something out of hand but will simply say 'could not be verified'. By your yardstick, all the scribblings of Ramanujam should be dismissed as 'abstruse'. Isn't it a fact that mathematicians are struggling with his scribblings? Therefore, let us not throw mud on our own faces.

BTW, is changing Adhara Shadjam applicable only to music? When we conduct long chanting sessions, don't we change it at times?