Making loops out of phonemes

First, thank you for making Audacity freely available.

Maybe you folks can give me helpful info on the following problem:

I recorded (German) words like “Katze” (cat) and “Chamäleon” (chameleon) with the Text-to-Speech service on The samples sound fine. (I used the voice “German - Julia”).

With Audacity 1.2.6, I cut the sample “Katze” into the individual sounds, i.e. “K”, “a”, “tz”, “e”, and saved them as individual WAV-files.

I wanted to make loops out of these individual sounds. I took a part from the middle of the “a” sample, put it in a new Audacity project and repeated it. But the result sounds very unnatural, like a very fast “Aa-Aa-Aa-Aa-Aa-Aa” (without gaps, though), not a continuous “aaaaaa”.
I tried to match the beginning and end of the selection which I looped. I put the beginning of a selection where the signal curve is at the bottom and starts to rise, and did the same for the end for the selection. Didn’t help, though.

(In the case of “tz”, I intended to make a looping “z” sound (omitting the “t”). But I had the same problem: the result sounded like a very fast “Zz-Zz-Zz-Zz-Zz”, not like “zzzzz”.)

Do you think it’s possible to take short word samples from that Text-to-Speech service and make decent loop sounds for the individual sounds of the word? If yes, could you give me any links to instructions for this on the net?
Or is it absolutely necessary to make much longer samples myself, i.e. saying “tzzzzzz” into a microphone and using the long “zzzzzz” to make a loop?
I really wanted to use that TTS synthesizer, though, as that’d enable me to make new sounds with their female voice anytime, instead of having to organize recording sessions every time.

Thanks in advance, Kibbi

Anything you do that way will probably always sound a bit “unnatural”…

One trick for making loops smoother is to use 2 samples and fade one out while you fade the other in. But in this case I don’t think you’ll be able to achieve what you’re looking for… but good luck anyway :slight_smile:

You could pick a very small sample in the middle of the ‘a’. Zoom a lot so you can see something similar to sine waves and find a very small portion (can be as small as 20ms) where the shape of the wave seems to repeat as a pattern. Cut in two points where what’s before and after those points seem to be very similar. Cut where the line crosses 0. Paste that sample into a new track and use the repeat effect and repeat it a few dozen times. You’ll probably get a somewhat continuous sound but probably sounding very “robotic”.

Hire a singer. It will probably be cheaper and less time consuming :wink: (and with much better results!)

Paul stretch” (free) will do the aaaaaah eeeee stuff: example by moi here …

“Paul stretch” also has a freeze option which makes the sound go on indefinitely rather than just stretched.

Audacity has “change tempo without changing pitch”, which in principle is the same as Paul stretch, but in practice it has more artifacts when taken to extremes, (and won’t do freeze).