Our websites:

Our social media sites - YouTube, Twitter and Facebook

YouTube X Facebook Instagram
731

What voice banking options are currently available?

Posted by Joanna Courtney on the 17th December, 2015

I read an interesting article in the RCSLT Bulletin today about the use of voice banking for patients with Motor Neurone Disease (MND) as a ‘Vocal Insurance’ policy.

The main message from the article is that voice-banking or ‘cloning’ needs to be ‘done early’ so that as much of the person’s own voice as possible can be used to create a better quality synthetic one. Even a non-fatiguing ‘healthy speaker,’ can find the process of recording their own voice time consuming and tiring. 

So what ‘Vocal insurance’ options are out there?

There are various options available from just over 1 hour’s worth of recording (about 400 sentences) up to around 8 hours of recording (1600 + sentences). Generally, the more sentences that are recorded the better the voice quality and likeness to the person’s own voice the synthetic voice is going to be.

The article talks about ModelTalker, which is being used in conjunction with Therapy Box communication app Predictable. ModelTalker is a free service, but you need to use the high quality voice created with the Predictable or Chatable app on iOS devices (iPads, iPhone etc). You can also get a Windows or Mac version and an Android version too. There is no cost, but they suggest a donation to the Nemours Speech Research Lab to support their research. The recording process is based on 1600 sentences and takes around 6 hours, depending on the individual. This does not have to be done ‘in one go’ and can be split over several days. There is a demo of the voice produced.

CereProc provide a voice-cloning service called CereProc me, which costs £499.99 to produce a high quality voice provided in Windows Sapi format in a couple of hours recording time. You can then use your voice on the Windows device of your choice. You can hear samples of the original recording and the synthetic voice produced on the website too.

The Voice Banking Project team, based at the Euan MacDonald Centre for MND Research are piloting a new voice banking system in Scotland, which uses ‘donor’ regional accent voices which are then ‘voice mixed’  with the patient’s own voice to create an ‘Average Voice Model’ synthetic voice based on just 400 sentences recorded in under an hour.

Acapela, who also make other text to speech voices, have a service called ‘My Own Voice’  It is free to record and play back the voice you create, but you need to pay varying amounts to use the voice in Windows, Android or in a particular application. There are 3 different levels of voice ‘quality,’ depending on the amount of recording data provided. The voices themselves are of a high quality when heard alongside the original recordings. The minimum number of sentences required to make a recording is about 1600, which may take between 5 and 8 hours to record the complete set. This can be split over several recording sessions if need be.

Each of these facilities take a slightly different approach and produce different ‘products’ which can be used on a variety of platforms and thus have varying costs. So, it’s best to find out more from the organisations directly before deciding on the option which suits you best.

Also remember that the three ‘ready-made’ synthetic Text to Speech Scottish voices;

Are available for FREE download for the Scottish Public Sector, distributed by CALL Scotland and funded by the Scottish Government. Over 10,000 downloads in the last 8 years.

Now, we just need some Scottish Children’s voices...

 

 

Online course - £30

Using AI to Support Learners with Dyslexia

Newsletter: join thousands of other people

Once a month we'll send you an email with news, research and thoughts, as well as training courses and free webinars you may wish to attend.