Acapela Provides Free and Quick Personal Voice Backup Facility for Everyone
Despite the apprehension surrounding synthetic speech in conjunction with deep fakes and AI trickeries, it remains a vital resource for individuals who have lost their ability to speak. Acapela Group’s latest offering, a free service named “my own voice,” is designed specifically for such individuals, allowing them to create their own AI voice profile.
With a quarter-century of experience in the text-to-speech domain, Acapela was recently purchased by the tech accessibility powerhouse, Tobii Dynavox. However, Acapela continues to operate autonomously.
The field of accessibility, like many others, has been significantly reshaped by the introduction of machine-learning processes at the consumer level. Remy Cadic, the co-founder of Acapela, recalls how just 7 or 8 years ago, personalizing a synthetic voice was not only a laborious task, but the output was far from satisfactory.
“In the past, patients had to undergo a lengthy process, requiring around 8 hours of training. Nowadays, we can archive a voice by recording just 50 sentences; this process takes about 10 minutes and the voice is ready by the following day,” he explained. “The evolution of neural text-to-speech techniques has certainly revolutionized the industry.”
The ability to utilize a speech generator featuring one’s own voice is undeniably becoming more appreciated. Selecting from a pre-set list can feel impersonal and somewhat strip away the human touch. Many individuals would prefer to use their own voices, but it’s only recently that such an option has become available.
Their claim about the process’s simplicity and speed holds true. I personally experienced the new “my own voice” procedure and it indeed involved reading just 50 brief sentences, seemingly randomly selected from a range of sources including novels, cookbooks, and articles. The recording interface was user-friendly and easy to use, and just as promised, my voice was ready for use in about a day. The quality was decent — not eerily lifelike as some models out there can be, but unmistakably my own voice, as promised. It managed to articulate any sentence I tested on the demo page.
Now that my voice has been stored, I can access and download it for a nominal charge, should I ever need it. It’s compatible with any appropriate speech-generation system. Naturally, this includes Tobii Dynavox’s TD Talk and devices. The company just launched a new product last week — these devices are becoming increasingly sophisticated and stylish.
The essence of all this isn’t to showcase the prowess of neural voice technology or to create a platform for cloning celebrity voices. Rather, it’s designed as a tool for those who, until recently, may have had no alternatives, or at best, would have to undergo a challenging and intricate process if they wished to retain their voice.
Individuals dealing with degenerative diseases, cancer, or certain medical procedures understand that their ability to speak may deteriorate or even cease entirely within a few months or years. Simplifying the process of preserving their voice is a service that many will undoubtedly appreciate.
“One significant advantage is that we also cater to children — we’ve made the recording script easier to comprehend and adjusted the system to improve the quality of synthetic children’s voices. We were the pioneers globally in this aspect, and we continue to progress in this direction,” Cadic stated.
The ability to record, re-record, or artificially age the preserved voice represents a novel and demanding capability, but it appears to be yielding positive outcomes:
A standout feature is its compatibility with offline devices that aren’t equipped with the most recent neural processing chips. “There are online solutions that facilitate easy voice creation, but they are solely accessible through the cloud, which isn’t always practical,” he explained.
While the 50-sentence process is beneficial for those who can still read and speak, a voice can also be trained using voice recordings from people who have since lost that capability — although the process isn’t as straightforward.
The company has also discovered the importance of diversity and careful consideration during the training process, similar to other AI applications. Cadic noted that one drawback of some ultra-fast training techniques is that “they essentially attempt to match the user with the closest speaker in the training material. However, if there isn’t a speaker in the training who closely resembles the original voice, the output simply won’t sound like it.”
Acapela’s product manager, Nicolas Mazars, further explained that, like many AI challenges rooted in inadequate training data, this issue doesn’t affect everyone equally: “The current process functions well for a typical 50-year-old Caucasian male, but not so for an African-American man, or for those with limited English proficiency. We operate in 23 languages and serve many users who live with disabilities. Our strategy is to depend on user feedback to develop a product for them, crafted by them.”
The procedure of recording and storing the voice is complimentary; you can register for an account here and prepare your own synthetic voice in a matter of minutes. The only time you need to pay is if you decide to download and implement it on a device.
Click here for the Latest update on Tech News.