Re JFK video: hear Kennedy's 'lost' Dallas speech in his own voice

["Dave Farber" <farber () gmail com>]

Begin forwarded message:

> From: "Wendy M. Grossman" <wendyg () pelicancrossing net>
> Date: March 25, 2018 at 9:20:16 AM EDT
> To: dave () farber net
> Subject: Re: [IP] JFK video: hear Kennedy's 'lost' Dallas speech in his own voice
>
> For IP if you wish:
>
> This story I wrote for Scientific American goes into more detail about
> how the same company made a synthetic voice for Roger Ebert:
>
> http://www.scientificamerican.com/article.cfm?id=new-speech-synthesis-could-make-robert-ebert-sound-more-like-himself
>
> wg
>
>
> > On 2018-03-25 13:58, Dave Farber wrote:
> >
> >
> >
> > Begin forwarded message:
> >
> > > *From:* Dewayne Hendricks <dewayne () warpspeed com
> > > <mailto:dewayne () warpspeed com>>
> > > *Date:* March 25, 2018 at 8:12:03 AM EDT
> > > *To:* Multiple recipients of Dewayne-Net <dewayne-net () warpspeed com
> > > <mailto:dewayne-net () warpspeed com>>
> > > *Subject:* *[Dewayne-Net] JFK video: hear Kennedy's 'lost' Dallas
> > > speech in his own voice*
> > > *Reply-To:* dewayne-net () warpspeed com <mailto:dewayne-net () warpspeed com>
> > >
> > > JFK video: hear Kennedy’s ‘lost’ Dallas speech in his own voice
> > > On November 22, 1963 John F Kennedy was silenced. Fifty-five years
> > > later we have made it possible for him to be heard again
> > > By Aaron Rogan
> > > Mar 15 2018
> > > <https://www.thetimes.co.uk/edition/news/jfk-video-hear-kennedys-lost-dallas-speech-in-his-own-voice-xtkvhm255>
> > >
> > > It took eight weeks to bring to life the 2,590 words John F Kennedy
> > > never got to speak.
> > >
> > > Sound engineers pulled 116,777 sound units from 831 of his speeches
> > > and radio addresses. These units were then split in half and analysed
> > > for pitch and energy. The half units, known as phones, were each about
> > > 0.04 seconds long and had to be tested next to each other to ensure
> > > that they did not clash. The W sound in weapons, for example, is not
> > > the same as the W sound in words.
> > >
> > > “There are only 40/45 phones in English so once you’ve got that set
> > > you can generate any word in the English language. The problem is that
> > > it would not sound natural because one sound merges into the sound
> > > next to it so they’re not really independent. You really need the
> > > sounds in the context of every other sound and that makes the database
> > > big,” Chris Pidcock, co-founder and chief voice engineer at CereProc
> > > in Edinburgh, said.
> > >
> > > How the speech was recreated
> > > Mr Pidcock’s company specialises in text-to-voice technology. It is
> > > used by brands to bring characters and products to life and also to
> > > allow people who are losing the power of speech from motor neurone
> > > disease or other conditions to maintain their own voice, something
> > > that is also part of their character.
> > >
> > > He has previously recreated Roger Ebert’s voice, at a time when the
> > > late American film critic had lost his speech because of cancer. That
> > > project involved trawling hundreds of hours of film commentaries from
> > > DVDs.
> > >
> > > Recreating Kennedy’s voice and unique cadence for the JFK Unsilenced
> > > project, to produce the speech he was due to give on the day he was
> > > assassinated in Dallas in 1963, proved a more difficult challenge.
> > > “Because of the old analogue recording devices used, it appeared as if
> > > it was a different person speaking each time. Trying to harmonise the
> > > environment and manipulate the audio so that it ran together was quite
> > > difficult,” Mr Pidcock said.
> > >
> > > He estimated that of the final 20-minute speech, fewer than half the
> > > audio units were beside each other when they were pulled from recordings.
> > >
> > > “Getting to that point is pretty challenging based on the variable
> > > audio quality, as well as the speech itself having different qualities
> > > and noise levels. One of the things we needed to do is get a very
> > > accurate transcription of the audio so that things like ‘umms’ and
> > > ‘ehs’ could be labelled and we could make sure the phonetic pieces we
> > > got were correct. If you label them incorrectly you might pick the
> > > wrong piece and the whole sentence will sound wrong.”
> > >
> > > The project involved a painstaking two-month process to pick the
> > > best-quality recordings of Kennedy’s voice and cross reference them
> > > with the written speech so that any crosstalk or static was not selected.
> > >
> > > Once a database of the cleanest sound units was built, a new computer
> > > system was employed to recognise and recreate Kennedy’s oratorical
> > > style. This required feeding data from his speeches into a computer
> > > until it learnt the patterns in his delivery. Then, when the speech
> > > was put in, the system could tweak it and make it more natural.
> > >
> > > [snip]
> > >
> > > Dewayne-Net RSS Feed: http://dewaynenet.wordpress.com/feed/
> > > Twitter: https://twitter.com/wa8dzp
> > Archives <https://www.listbox.com/member/archive/247/=now> | Modify
> > <https://www.listbox.com/member/?&;>
> > Your Subscription | Unsubscribe Now
> > <https://www.listbox.com/unsubscribe/?&&post_id=20180325085847:3FC8451C-302C-11E8-9483-C8555F67D5ED>
> >    [Powered by Listbox] <http://www.listbox.com>
>
> -- 
> www.pelicancrossing.net <-- all about me
> Twitter: @wendyg



-------------------------------------------
Archives: https://www.listbox.com/member/archive/247/=now
Modify Your Subscription: https://www.listbox.com/member/?member_id=18849915&id_secret=18849915-aa268125
Unsubscribe Now: 
https://www.listbox.com/unsubscribe/?member_id=18849915&id_secret=18849915-32545cb4&post_id=20180325111241:F49937F0-303E-11E8-82C8-D8C866047D21
Powered by Listbox: http://www.listbox.com