Female Robotic Text-To-Speech Tutorial

[wolterh]

Hi, I've been trying to get a robotic voice TTS to replace the default voice in espeak, which is boring and terrible.

I have achieved a female robotic voice, which I explain how to obtain here.

The tools used are

espeak
sox (play)


First, to "say" something with espeak, the TTS engine, you can use the following code:

echo "All systems online" | espeak

The voice the text is read with is the one we want to change. Specifically for the female robotic voice, we are going to use as base voice the female3 voice. We can set espeak to use it with the -v parameter, followed by the name of the voice: e.g.

echo "I sound like a female" | espeak -v female3

Now is when the robotic part comes in. Robotic voices are typically characterized by two things:

Echo
Metallic sound


These effects we can achieve with the sound manipulation application play, which is an interface to sox used solely for playing audio.
You can read the manual on sox
man sox to understand the following code, which will transform normal computer voice into robotic-ish voice. You should read the espeak manual as well.


OBTAINING THE ROBOTIC VOICE
To get the robotic voice, follow these steps:

Save the following text in a text file (in this example, it will be referred to as femalebot:


espeak --stdout -v female3 | play -t wav - \
overdrive 10 \
echo 0.8 0.8 5 0.7 \
echo 0.8 0.7 6 0.7 \
echo 0.8 0.7 10 0.7 \
echo 0.8 0.7 12 0.7 \
echo 0.8 0.88 12 0.7 \
echo 0.8 0.88 30 0.7 \
echo 0.6 0.6 60 0.7 \
gain 8

The metallic effect is achieved by using the echo effect with a very short delay. The overdrive enhances the sound a bit, and the other echo effects give the final kick to the robotic voice.


Now, set it to be executable with the chmod command, e.g.

chmod +x /path/to/femalebot

You can now use the robot voice in TTS, by issuing the following command:


echo "All systems online." | /absolute/path/to/femalebot



Additionally, and to make things easier, you can place the script in any directory in the PATH variable. This way, you can issue the last command as follows:


echo "All systems online." | femalebot

[coolglobal]

Sweet! Very sci-fi! Works as described. I'm going to have fun with this.

[Bandit]

Very nice.
I have been playing around with it every now and then for a month or two trying to get the female voice to sound more life like.

[Aquix]

Nice, tts is fun to play with. I'll test it this weekend.

Btw, it's called fembot ;P

[lovinglinux]

Cool. Now I can add voice to my scripts. :-)

BTW, the sox script didn't work for me.

[tjeremiah]

ill give this a shot. I tried this on KDE because they have this feature where it can tell you the time, but I couldnt get it to sound normal.

[wolterh]

You can easily let it tell you the time by adding a cron job :) I did it myself, on an hourly basis.

However, I did some modifications to my script, I think it sounds a bit better:



espeak --stdout -v female3 | play -t wav - \
overdrive 10 \
echo 0.8 0.8 5 0.7 \
echo 0.8 0.7 6 0.7 \
echo 0.8 0.7 10 0.7 \
echo 0.8 0.7 12 0.7 \
echo 0.8 0.88 12 0.7 \
echo 0.8 0.88 30 0.7 \
echo 0.6 0.6 60 0.7 \
gain 8



Cool. Now I can add voice to my scripts. :-)

BTW, the sox script didn't work for me.

Does it throw an error or something?

[uRock]

Moved to Tutorials & Tips. In the future please post your tutorials here and wait for them to be approved.

Thanks,
uRock

[Bandit]

Here is one I think is fairly good. Its less glass sounding then Wolis but will require some minor editing to one of the voices in the espeak shared data directory.

Editing the F5 voice in the voices/!v folder, copy and past this data over the existing voice info in that file:



language variant
name female5
gender female

pitch 165 220
roughness 0

formant 0 105 80 150
formant 1 110 80 160
formant 2 110 70 150
formant 3 110 70 150
formant 4 115 80 200
formant 5 115 80 100
formant 6 110 70 150
formant 7 110 70 100
formant 8 110 70 150

stressAdd 0 0 -10 -10 0 0 10 40
breath 5 14 16 16 16 6 0 10
echo 20 10
voicing 75
consonants 150 150
breathw 150 150 200 200 400 400


Then use this script to get the voice fine tuned a great deal:

espeak --stdout -s120 -k18 -a200 -v female5 | play -t wav - \
chorus 0.4 0.8 20 0.5 0.10 2 -t \
echo 0.9 0.8 33 0.9 \
echo 0.7 0.7 10 0.2 \
echo 0.9 0.2 55 0.5 \
gain 10

Cheers..

[wolterh]

Nice job! The female voice is a lot more clear! Sometimes I barely understand the other one!

[Bandit]

Nice job! The female voice is a lot more clear! Sometimes I barely understand the other one!

Thanks, I think I can get it more clear. But what it boils down to is adding a lot of stops and pauses in the wording to make it sound more natural.

[H3g3m0n]

Been looking for a way to make voices sound more robotic myself.

Those scripts are nice but the problem is they all have those weird accents, they sound like a combination of mentally handicapped, deaf and European (even the US voices).

Might be worth looking at trying to use OpenMARY (http://mary.dfki.de/) instead of espeak. Has some more natural base voices. And running that through the overdrive.

Current speech synthesis just sounds so bad, even the top end stuff. Even if they do manage to get it sounding right you will never have the emotional tones that get added in when talking about different topics.

It's better to just avoid the uncanny valley by not trying to make things sound as real as possible.

Also I don't see why I would want my computer to sound like a person when it's not. I would prefer it to be obviously a computer.

[wolterh]

OpenMARY sounds so much better than the espeak voices! I think the robotic voice will sound a lot less cryptic

[go_beep_yourself]

Have you taken a look at JSpeak?

http://linuxinnovations.blogspot.com/2012/05/jspeak-ultimate-in-linux-text-to-speach.html