Finding a Voice

About a week ago I posted about the possibility of introducing an animated news reader character to The Sleaze. Well, things have moved pretty quickly since then, and not only have I created such a character, but I've pretty much finished trialling it off-line. Indeed, if I find time, I hope to deploy her some when in the next few days. Actually, creating the actual character was the - relatively - easy part. I used the Sitepal service. As a Bravenet user, I was able to create a free account with Sitepal, (I don't know if this offer is still valid - it's worth checking if you are interested in this sort of thing), and customise one of their existing animated avatars. You can change just about any aspect of these - eye colour, hair, skin tone, even age - to create a unique character. It didn't take me long to come up with a character I liked, the problem was coming up with a voice for her.

The only real catch with using Sitepal's free basic account is that the speech options are limited. The character can speak a standard, generic, message from the site's library, alternatively, you can record your own voice, (not really an option, as I'm male and the character is female). Whilst there is a text-to-speech (TTS) option on the control panel, (with a variety of voices), it turns out that this is only available on paid accounts. So, I only had one alternative - to record a script externally, and upload it as an mp3 file for the avatar to read. Now, whilst there are a number of free text-to-speech services available, (both online and and as downloads), most use the very outdated and robotic SAPI4 voices. Whilst there are plenty of the somewhat superior SAPI5 synthetic voices out there on the web, all are commercial and require licencing fees. Moreover, most of the UK accented voices are, frankly, terrible. Not their accents, but their pronunciations of many words is downright peculiar. The result, I assume of them having been developed by Americans who have some strange ideas about the way we Brits speak.

This left me with one realistic option - find a TTS converter that used my laptop's native SAPI5 voice: Microsoft Anna. Now, Anna has an American accent, pronounces some words oddly, and isn't quite as smooth as some of the commercially available voices, but she was available to me free. Which was important, as I was determined that this project wasn't going to cost me anything other than time. Anyway, I eventually found a piece of Russian TTS software called Balabolka, which not only used Anna as its default voice and produced mp3 files, but also allows the user to adjust the pauses between sentences and paragraphs so as to produce a more human-sounding speech rhythm. To cut a long story short, I've now managed to record an introductory script for my character and have successfully uploaded it. The result is actually not at all bad. Sure, the voice is still a bit metallic in places and some of her pronunciations are a little eccentric, but it is all clearly audible and the character lip-synchs to it quite effectively. So, you can expect to see Suzie Sleaze, (I used the name for the voice tests and it stuck), reading the showbiz news soon.