‘Voice banking’ has been discussed quite a lot within the AAC field recently and so, as a team, we have been exploring this and other similar techniques in more depth.
As the team member volunteered to test out the packages available to create a personalised synthetic voice I have spent what feels like weeks recording countless phrases! I have now created two personalised voices, one using ModelTalker and one using My-Own-Voice. This post is a summary of the experience of creating these voices (including example recordings of the result). We also have a fact sheet on our website with information about the options.
Last week it was fantastic to finally hear the results of all the hours of recording into the ModelTalker voice recorder. The process has been frustrating to say the least! I have good quality voice, had access to the quiet surroundings required and a decent microphone and was pretty competent in the IT stuff needed to set up the ModelTalker recordings – but even so it wasn’t plain sailing. At times it seemed impossible to get a “green” light recording with all parameters at an acceptable level, and so, individual phrases had to be re-recorded numerous times, even though as far as I was aware all environmental factors, recording volume and voice quality were unchanged. The tiring factor, even for me with a robust voice, was telling, and it was difficult to record successfully for more than a couple of hours at a time. Connection at times was lost, and at times the programme froze, which again slowed the recording process and sometimes meant that the time set aside to record could not be used.
The ModelTalker documentation suggests that the recordings can be completed in 6-8 hours. However, it took me far longer than this and a lot of my phrases were submitted as an “amber” recording; meaning acceptable, but not perfect. Consequently, it is very evident that for most of our clients a comprehensive level of support during the recording process would probably be needed to get the best outcome.
However, on hearing the final product I can safely say that the voice that has been created definitely sounds like me – and creating and using this voice is currently completely free! The speech produced can be a little disjointed at times, irregular English spelling patterns create it some pronunciation difficulties and the intelligibility decreases in longer utterances, but the overall quality definitely shows features of my voice
Alongside this process I have begun working with a client, Greg, who, with support from his local therapist, Jennifer Benson, has created his own ModelTalker voice which he now uses on Predictable. To hear about the process I had just undertaken, from a client’s perspective, has been fascinating . Greg came across similar frustrations to those I have highlighted during the recording process and the tiring effect was very significant for him. However, for Greg, the pay off of having his own voice on his communication app seems to outweigh any of the difficulties he encountered.in creating his voice. Greg has made a video about his voice banking experience at: https://youtu.be/DYdSTNDYBWE.
After the hours taken to record my voice using the ModelTalker platform, the creation of a personalised voice using “My-Own-Voice” seemed a lot less time consuming. Although the total number of phrases needed to be recorded were similar to ModelTalker, the “My-Own-Voice” only took around 5 hours to make. The re-recordings needed were minimal and generally the process was less disrupted, with the recording working consistently each time I attempted to record a phrase. The navigation through the phrases as I recorded them seemed a lot more intuitive and I was able to seamlessly record one phrase after another.
Once more the voice created does sound like me, although there are some definite issues! Certain speech sounds do not sound at all like they should, particularly word endings and some word initial consonant blends. In connected speech, intonation patterns can sound a little odd at times and the boundaries between words sound quite slurred which definitely has a negative impact on intelligibility.
The My-Own-Voice process is free to record and create a voice, but you then need to apply for a costing to use the voice you have created on a communication aid.
Comparing the Results
For the purposes of comparison, I recorded a phrase as a direct voice recording and then created it using my personalised synthetic speech with “ModelTalker” and “My-Own-Voice”. You can compare the results for yourself by listening below, the phrase I used was “I am sat here, writing this blog, to allow you to compare the personalised voices that can be created by two web based programmes; ModelTalker and My-Own-Voice”:
My recorded voice
ModelTalker synthesised speech.
My-Own-Voice synthesised speech.
Feel free to download the information sheet we have produced about voice and message banking from our website. This summary includes some hints and tips to think about when considering the process of recording words and/or phrases or creating a synthetic personalised voice.
In a future post we will discuss the difference between Voice Banking, Message Banking and other approaches to retaining identity in the use of communication aids. As we write this post, breaking news is that Amy Roman, an AAC specialist in the USA has created a resource for message banking – MessageBanking.Com . This is discussed on this thread on the fantastic AT ALS email list.