Having Trouble Hearing The Tones In Thai Words? I Have Made A Short Video On How You Can Use Free Software To Visualise The Tones

[jamiesensei]

Hi all,

I posted a screen cast on my web site that describes how to use the completely free software 'Audacity' to see a spectrum graph of the tones in Thai words! See the video here : http://jamiep.org/mod/resource/view.php?id=207

You can clearly see the different tones in sound files that you record yourself or that you load into Audacity. See for example the screen shot attached. The screen shot is a spectrum graph for the word 'ma' pronounced with the five different tones - mid tone, low tone, falling, high and rising.

I am real excited about this as it means I can record my own voice and then compare it with a recording of a Thai native speakers voice.

Jamie

[kriswillems]

Nice demo Jamie.

There are also open source real time spectrum analyzers on the internet.

These are a bit easier to use because you get immediate feedback when you speaking.

For instance:

http://www.speech.kth.se/wavesurfer/index.html

The software is BSD licensed and available for MAC, linux, windows, ...

This software is very powerful for speech analysis. It would be nice to have a demo for this one.

Before many people used the software on slice-of-thai. But that one is windows only and not open source.

Edited June 18, 2009 by kriswillems

[jamiesensei]

For instance:

http://www.speech.kth.se/wavesurfer/index.html

Thanks Kris. The above software seemed a bit buggy at first on Windows Vista but seems to be running OK now and does look to have more useful features than Audacity for this kind of thing.

[katana]

Interesting. Didn't know Audacity could display frequency-time graphs like that for the Thai tones.

[RickBradford]

Many thanks for that, Jamie.

A very comprehensive and clear demo, although your style is too relaxed to get you a job presenting at IBM....

[kriswillems]

When analyzing tones it's a good idea to look at the first harmonic only. (A harmonic matches with a fat line in the spectrogram, so you would have to look at the lowest line). This line is somewhere around 100 Hz (for mid tone or 150Hz for high tone).

If you look at the first harmonic you can compare the tone height more easily.

So the maximum frequency in your spectogram should be set around 300 Hz for men of 400 Hz for women.

It's a good idea the set your sample frequency not too high, maybe only at 8 KHz. This way you can filter out some noise caused by a bad sound card, bad microphone or bad recording environnement.

Settings for Wavesurfer are:

- right click in gray area -create pane -spectogram

- right click in the spectogram - properties

- In the spectogram TAB-sheet:

FFT window size 4096 point

Analysis bandwidth 10 Hz

Cut of spectogram at 300 Hz or 400Hz

Record scroll speed 50 pixels per second

- In the sound TAB-sheet:

sample rate 8000 Hz

Click Apply

Click OK

- Click on the second button from the right in button-bar to zoom time scale to windows size.

Now you get record and get a real time analysis.

(PS. you can save your configuration so you don't need to set everything every time you start wavesurfer).

picture attached.

[meadish_sweetball]

Some very useful tips in this thread. Thanks to all who have participated.

When analyzing tones it's a good idea to look at the first harmonic only. (A harmonic matches with a fat line in the spectrogram, so you would have to look at the lowest line). This line is somewhere around 100 Hz (for mid tone or 150Hz for high tone).

Do you know why the harmonic is more indicative? Is it because it is more sharply defined and less likely to be subject to interference?

[kriswillems]

The first harmonic is the base sound of your voice. It's the sound without "color". It's a pure sine wave as you would get from a tuning-fork. The second harmonic is also a sine wave which is super-imposed on the first harmonic. Usually it amplitude is smaller and therefore it's less important. The second harmonic has exactly the double frequency of the first harmonic. There are more harmonic waves that add more color to the sound.

The reason why it's best to compare the first harmonic is that you don't need to count.

If you would for instance use the third harmonic you would need to compare the third harmonic of the low tone with the third harmonic the middle, rising, high and falling tone. This would involve a lot of counting.

If you compare the first harmonic it's very easy to see the low tone has a lower frequency than the mid tone. And it's very easy to see the high tone has a higher frequency that the mid tone.

Another reason to use the lower harmonics is that they have the biggest amplitude and are less sensitive for interference.

An ideal tone analyzer (for practicing the tones in Thai language) would filter out the other harmonic tones via a low pass filter, so you get a single line graph. I didn't find any open source real time spectrum analyzer that can do that. But it would be not to difficult to adapt existing open source software.

Edited June 21, 2009 by kriswillems

[jamiesensei]

Hi Kris,

Do you notice on the screen shot I attached to my original post that it is easier to see the change in tone during the pronunciation of the syllable in the upper harmonics? The first harmonic's tone is almost flat but the upper harmonics tone does change considerably during the pronunciation of the syllable!

I really don't know but I would guess that the ear is more sensitive to tones above 400 Hz?? So although on the spectogram the harmonics appear to me to have almost equal power (although I could be interpreting the graph wrong), I would guess the harmonics around 1-2.5 KHz would be more important in terms of determining what someone actually hears, as this would be a range that the ear is more sensitive too??

However I agree that if you want to determine the relative tone of each syllable and not how the tone changes during the syllable looking at the first harmonic makes sense.

Do you think my reasoning makes sense?

[kriswillems]

Hi Kris,

Do you notice on the screen shot I attached to my original post that it is easier to see the change in tone during the pronunciation of the syllable in the upper harmonics? The first harmonic's tone is almost flat but the upper harmonics tone does change considerably during the pronunciation of the syllable!

I really don't know but I would guess that the ear is more sensitive to tones above 400 Hz?? So although on the spectogram the harmonics appear to me to have almost equal power (although I could be interpreting the graph wrong), I would guess the harmonics around 1-2.5 KHz would be more important in terms of determining what someone actually hears, as this would be a range that the ear is more sensitive too??

However I agree that if you want to determine the relative tone of each syllable and not how the tone changes during the syllable looking at the first harmonic makes sense.

Do you think my reasoning makes sense?

Yes jamie, that makes sense. The absolute tone change in Hz is bigger in the upper harmonics. I can't explain why. Does somebody know?

But if you change the scale of your graph and zoom in on the lower harmonics you would notice that the relative change in tone is also big in the lower harmonics.

100Hz -> 150Hz = 50% change, but only 50 Hz

1000Hz ->1500Hz = 50% change, but 500 Hz

So, in the end it's just a matter of which scale you use.

I also think it's not important to show the harmonics the ear can actually hear best. You just want to have a good visual representation of the tone change and you want to be able to compare your different tones when you pronounce them. 1000Hz would be the 8th to 10th harmonic... that's a lot of counting. If we would have software to do the counting for us comparing the 8th to 10th harmonic would be a good alternative.

Edited June 22, 2009 by kriswillems

[RickBradford]

^

The absolute tone change in Hz is bigger in the upper harmonics. I can't explain why. Does somebody know?

Frequencies are geometric not linear. Every octave higher indicates a doubling in frequency i.e. the frequencies go 1-2-4-8-16, not 1-2-3-4-5. So there will inevitably be a higher absolute Hz difference at the high end. Is there perhaps a setting in the software called 'exponential' or 'geometric'?