Slashdot Mirror
Star Trek's LCARS Could Become Your Virtual Assistant (cnet.com)
H_Fisher writes: It has arguably inspired many other technological innovations in the fifty years since its premiere, and now another Star Trek-inspired touch could be coming to your device: the voice of Majel Barrett from the Star Trek universe's LCARS computer system. CNET reports: "The voice of LCARS was provided by Majel Barrett, who was married to Star Trek creator Gene Roddenberry. Although Barrett sadly passed away in 2008, she took several roles on the show over the years, including nurse Christine Chapel in Star Trek: The Original Series and Betazoid ambassador Lwaxana Troi on Star Trek: The Next Generation. According to a tweet by the official Roddenberry account yesterday, this has provided enough phonetic data to perhaps get Barrett's voice appearing in upcoming new 2017 TV series Star Trek: Discovery -- and maybe even a Siri-like virtual assistant."
They could do that today. They did it with Hitler for a documentary by digitally putting his face into an actor.
I kinda wish they would actually. Bring back Firefly, finish Angel, do some more Next Generation stories with the characters from that era.
const int one = 65536; (Silvermoon, Texture.cs)
SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
That's how the synthetic voices of Siri and Cortana are made, too. Voice actors read texts, the recordings are split up into phonemes and these are then used to synthesize the actual words we can hear. These computer voices aren't made on a word-by-word basis anymore.
The recorded text however are nonsense texts that are specially designed to contain a maximum phonem variety in a minimum of recorded text and that way of course it's known how many vairants of each phonem are available exactly where in the recording. So, with enough recorded material it should be possible to extract the same phonem variety. It's just more work as you have to find them first
bickerdyke
The number of actual complete words recorded can improve the overall quality of the synthesized voice. Phonetically pasted-together words are still not quite as good as complete, well formed words. (That's why they always have the talent read the full set of numbers instead of synthesizing words like "eleven".) Reading a phonetically complete subset of words is a good way to capture the most usable portion of the voice in the minimum amount of time. That's important when you're paying the talent by the hour, but it's not necessarily going to produce the overall best results. Having access to the full body of work will not only provide the needed phonemes, but will include a good vocabulary of higher quality words.
Of course, having a slightly choppy computer voice is one way of overcoming the uncanny valley. Holding a conversation with a dead person might be unnerving for some people. Hearing the little clips and weird tone changes as the voice is reassembled would be a constant reminder that you're actually talking to a computer, not a person, and might be of some comfort.
John