Author Topic: DaveClone  (Read 14337 times)

0 Members and 1 Guest are viewing this topic.

Offline MK14

  • Super Contributor
  • ***
  • Posts: 4703
  • Country: gb
Re: DaveClone
« Reply #25 on: May 08, 2023, 05:31:14 am »
If you go back a number of decades, the state of artifical voice technology, e.g. Stephen Hawking's, was:



So, compared to that, rather noticeable robotic, non-human voice.  It has come along, in leaps and bounds.

If, because of medical reasons, one had to lose ones voice for a few months.  I think the quality, is pretty good (as a temporary fix, until recovery from the medical situation).  Even if it is not bang on, in a number of respects, such as accent.
« Last Edit: May 08, 2023, 05:32:57 am by MK14 »
 

Offline chickenHeadKnob

  • Super Contributor
  • ***
  • Posts: 1059
  • Country: ca
Re: DaveClone
« Reply #26 on: May 08, 2023, 09:28:22 pm »
I hear the the AI voice as averaging out and flattening your accent. Much in the same way people, typically actors who spend equal time in England and North America,  are often accused of having a "mid" Atlantic accent.

There is a fellow Australian you-tuber, Mr Green who makes parody videos of nilered. The voice is quite good. short:
https://www.youtube.com/shorts/kBR4XuBH1AE

He describes his process in this video, skip to 15:16. He is somewhat coy as to the actual software he is using, there must be a non-disclosure clause.

 
The following users thanked this post: MK14

Offline MK14

  • Super Contributor
  • ***
  • Posts: 4703
  • Country: gb
Re: DaveClone
« Reply #27 on: May 09, 2023, 04:02:47 am »
My understanding (from memory and could be wrong), is that a person has difficulties, properly hearing (comparing) their own/real voice from recordings.  Because, unlike when other people speak, you hear your voice more directly and more via direct vibrations and things.
I.e. You are NOT (contrary to a persons opinion), necessarily the best independent judge of how real it sounds.

I have literally thousands of hours of training hearing my own voice through speakers exactly as you and everyone else has. Remember, when I edit my videos I'm listening through speakers, I'm not hearing my voice internally.

I think those thousands of hours would probably improve your ability to tell the quality of something trying to imitate your voice.  But there might still be residual psychological (biases) affects, because of the partially weird way, humans hear their own spoken words (speech).

[Hypothetically]
I suspect that if you phoned up some of the members of this forum.  But used the 'text to speech' voice translator (as you did, in some of your recent AI videos).  A number of members, would be fooled into thinking it was the real Dave Jones of EEVblog, voice talking.

On the other hand.  If you limited those 'test' phone calls, to only people who had already spoken to the real you by phone, on a huge number of occasions.  E.g. At least many hundreds of hours.  They would be considerably more likely to tell, it wasn't your real voice.

I.e. It perhaps is good enough to 'fool' people, less familiar with the real you and/or people who are not so good, at telling if a speaker is real or not (fake).
« Last Edit: May 09, 2023, 04:07:45 am by MK14 »
 

Offline Mr.B

  • Supporter
  • ****
  • Posts: 1245
  • Country: nz
Re: DaveClone
« Reply #28 on: May 09, 2023, 04:15:04 am »
Dave: What is your wife’s opinion of the fake voice?
She is probably the most qualified to rate its closeness to the real thing.
Where are we going, and why are we in a handbasket?
 
The following users thanked this post: thm_w, MK14, Jacon, golden_labels

Offline thm_w

  • Super Contributor
  • ***
  • Posts: 6707
  • Country: ca
  • Non-expert
Re: DaveClone
« Reply #29 on: May 09, 2023, 08:07:21 pm »
For the first video its interesting how most people are saying "8/10 its Dave but without the accent". For me the accent is a huge difference, so it doesn't sound like him at all.

Rough idea for my points scale might be:
- average pitch 2pt
- variation in pitch/intonations 2pt
- cadence 2pt
- accent 2pt+?
- other ?

Yeah its got the average pitch and cadence right, but, its missing the rest of it (accent, intonations). 4/10.
Profile -> Modify profile -> Look and Layout ->  Don't show users' signatures
 

Offline EEVblogTopic starter

  • Administrator
  • *****
  • Posts: 38039
  • Country: au
    • EEVblog
Re: DaveClone
« Reply #30 on: May 10, 2023, 03:09:55 am »
Dave: What is your wife’s opinion of the fake voice?
She is probably the most qualified to rate its closeness to the real thing.

Asked her last night.
Two words into CloneDave and she laughed and said pommy accent. She also rated it a complete fail because of the accent.
 
The following users thanked this post: thm_w, MK14

Offline EEVblogTopic starter

  • Administrator
  • *****
  • Posts: 38039
  • Country: au
    • EEVblog
Re: DaveClone
« Reply #31 on: May 10, 2023, 03:12:17 am »
My understanding (from memory and could be wrong), is that a person has difficulties, properly hearing (comparing) their own/real voice from recordings.  Because, unlike when other people speak, you hear your voice more directly and more via direct vibrations and things.
I.e. You are NOT (contrary to a persons opinion), necessarily the best independent judge of how real it sounds.

I have literally thousands of hours of training hearing my own voice through speakers exactly as you and everyone else has. Remember, when I edit my videos I'm listening through speakers, I'm not hearing my voice internally.

I think those thousands of hours would probably improve your ability to tell the quality of something trying to imitate your voice.  But there might still be residual psychological (biases) affects, because of the partially weird way, humans hear their own spoken words (speech).

Any "bias" would be because of where I'm from. I bet you there isn't a single aussie who would not identify the voice as clearly a british accent, and therefore rate it as a fail as a result.
 
The following users thanked this post: MK14

Offline MLXXXp

  • Frequent Contributor
  • **
  • Posts: 337
  • Country: ca
Re: DaveClone
« Reply #32 on: May 10, 2023, 06:13:39 pm »
An AI won't be able to believably sound like Dave unless it adds "and stuff like that", "and things like that" or a similar filler at the end of every 3rd or 4th sentence that it's given to speak.  ;)
 

Offline james_s

  • Super Contributor
  • ***
  • Posts: 21611
  • Country: us
Re: DaveClone
« Reply #33 on: May 10, 2023, 07:00:30 pm »
I'd say it sounds surprisingly close, like it could be Dave talking over the phone or recorded with a crappy microphone or something. It's not spot on, but it's probably 90% there and if someone told me that was Dave's voice I'd probably believe them.
 

Offline MK14

  • Super Contributor
  • ***
  • Posts: 4703
  • Country: gb
Re: DaveClone
« Reply #34 on: May 10, 2023, 10:54:44 pm »
Asked her last night.
Two words into CloneDave and she laughed and said pommy accent. She also rated it a complete fail because of the accent.

Any "bias" would be because of where I'm from. I bet you there isn't a single aussie who would not identify the voice as clearly a british accent, and therefore rate it as a fail as a result.

Yes, I own up, you have got me hook, line and sinker, here.

Ironically, I've fallen, for exactly the phenomenon, that I was suggesting, you might be suffering from (bias and possible lack of proper perception, of how the speech actually sounds like, because it was your own voice).

Because of my significant lack of experience with aussie accents, and I'm probably relatively (compared to the general/average population), not especially good with accents.  I wasn't hit by anything like the same, hint/signal/signs, when listening to the 'fake' Dave Jones, voice patterns.

[Wild Speculation]:  Because of the way the AI system was trained, with (presumably) lots of available voices, which (also presumably) were American and British accent voices, rather than aussie ones.  Perhaps that made the AI system you used to create the 'fake' voice, tend to create new voices, based very strongly on the training data (voices), which if they lack much/any aussie in them.

Which perhaps means it will tend to make 'fake' voices, with potentially the wrong accents, if it was missing or mostly absent, from the original training data.
But as I said, this is 'wild speculation', as I don't really know the specifics of how the system you used, was created and trained.
 

Offline thm_w

  • Super Contributor
  • ***
  • Posts: 6707
  • Country: ca
  • Non-expert
Re: DaveClone
« Reply #35 on: May 11, 2023, 12:28:08 am »
Been around many Australians so maybe thats why its easy to pick up. Just don't ask me to tell the difference between new zealand and australian accent.
Profile -> Modify profile -> Look and Layout ->  Don't show users' signatures
 
The following users thanked this post: MK14

Offline EEVblogTopic starter

  • Administrator
  • *****
  • Posts: 38039
  • Country: au
    • EEVblog
Re: DaveClone
« Reply #36 on: May 11, 2023, 02:04:35 am »
[Wild Speculation]:  Because of the way the AI system was trained, with (presumably) lots of available voices, which (also presumably) were American and British accent voices, rather than aussie ones.  Perhaps that made the AI system you used to create the 'fake' voice, tend to create new voices, based very strongly on the training data (voices), which if they lack much/any aussie in them.

No doubt.
 
The following users thanked this post: MK14

Online Zero999

  • Super Contributor
  • ***
  • Posts: 19676
  • Country: gb
  • 0999
Re: DaveClone
« Reply #37 on: May 12, 2023, 08:14:21 pm »
[Wild Speculation]:  Because of the way the AI system was trained, with (presumably) lots of available voices, which (also presumably) were American and British accent voices, rather than aussie ones.  Perhaps that made the AI system you used to create the 'fake' voice, tend to create new voices, based very strongly on the training data (voices), which if they lack much/any aussie in them.

No doubt.
Some would say that's discrimination. ;)

Is there anyone else with an unusual accent, but not close to standard British or American English, who's tried this? There's no point in me doing it, because my accent is fairly close to boring generic British. It would be interesting to see what it makes of a South African, or Scottish accent.
 
The following users thanked this post: MK14

Offline vk6zgo

  • Super Contributor
  • ***
  • Posts: 7653
  • Country: au
Re: DaveClone
« Reply #38 on: May 13, 2023, 01:04:20 am »
[Wild Speculation]:  Because of the way the AI system was trained, with (presumably) lots of available voices, which (also presumably) were American and British accent voices, rather than aussie ones.  Perhaps that made the AI system you used to create the 'fake' voice, tend to create new voices, based very strongly on the training data (voices), which if they lack much/any aussie in them.

No doubt.
Some would say that's discrimination. ;)

Is there anyone else with an unusual accent, but not close to standard British or American English, who's tried this? There's no point in me doing it, because my accent is fairly close to boring generic British. It would be interesting to see what it makes of a South African, or Scottish accent.

Of course, there are variations in the basic Aussie accent.
A lot of people who "identify" as Australians are migrants from elsewhere who have been in this country for many years.

English people in particular tend to develop an interesting "hybrid" accent over many years, which is still recognisably Australian, but is distinct enough for Australians to say "Ok, you are English by birth, but have spent a long time in Oz". *
(Other "Poms", on the other hand, either keep their original accent, or lose any vestiges of it, & speak with a normal Australian accent)

The late actor Trevor Goddard, although he only spent a few months in Perth WA years ago, had that accent "down pat", when he played ""Lt. Cmdr. Mick Brumby" RAN in the "JAG" TV series.
Hearing it, I immediately placed him in the above* category.

Mr Goddard showed the "knack" that some humans have for picking up nuances of accents, that it is "a bit of an ask" to expect AI to emulate at this quite early stage.

Back in the day, when artificial speech was making its very basic steps, many of its early iterations sounded very "Dalek like", but people would start to "tune in" to them mentally, making their speech more understandable.
I was quite happy with my adaptation to them, until I heard one speaking French!
 
The following users thanked this post: MK14

Offline EEVblogTopic starter

  • Administrator
  • *****
  • Posts: 38039
  • Country: au
    • EEVblog
Re: DaveClone
« Reply #39 on: May 14, 2023, 11:06:41 am »
It would be interesting to see what it makes of a South African, or Scottish accent.

I'd pay money to hear an AI version of Mark Meechan (Count Dankula) try and do jokes.
 
The following users thanked this post: MK14

Offline AndyBeez

  • Frequent Contributor
  • **
  • Posts: 856
  • Country: nu
Re: DaveClone
« Reply #40 on: May 14, 2023, 02:14:20 pm »
The UK is an accent soup. Not only can you tell from which city someone originates but also, from which side of that city. London I believe has at least 10 distinct accent groups? I will also add another London accent group which is a type of social media accent. This whiney accent afflicts 20 somethings who grew up watching the Carcrashians on UTube. It's common English mixed with west coast L.A. Makes the speaker sound double dim  :-\

On the subject of AI, PCs/Macs have system voices in many accents too. Test your localisation by asking the accented voice to speak various phrases. I found the Scottish accent on a Mac is highly adept at swearing - for some reason?!  :rant:Haggis!

Meanwhile, Apple's female system voice in an Aussie accent is called Karen. I have no idea why that is? See [Settings > Accessibility > Spoken Content > System Voice ] if you're interested in hearing "winner winner chicken dinner" uttered by a generic Cath and Kim.

Comparing computer accents might make an interesting segment on the AmpHour/etc  :wtf:

++ I'll add here the worse British English accent has to be the northern Irish or Ulster accent. Curiously, the southern Irish accent is one of the nicest. Weirdest is the impenitrable Newfoundland accent which is some mysterious blend of Gaelic, French and f-knows. As for the Kiwis, sorry to hear it's still legal to have six on a Sunday.
« Last Edit: May 14, 2023, 04:51:56 pm by AndyBeez »
 
The following users thanked this post: MK14

Offline RJSV

  • Super Contributor
  • ***
  • Posts: 2277
  • Country: us
Re: DaveClone
« Reply #41 on: May 15, 2023, 12:49:58 am »
   I'd give the BOT version a '7' and with my comment that I noticed, in the BOT's third sentence being read, that there was a noticable 'swell' or volume increase at the start of sentence.  That sounded very natural 'human' to modulate speech like that.
   All you 'Dave's out there are way too precise, knowing and comparing to the BOT version, but most people don't listen so critically.  To myself, that sounded human, and I wasn't looking for clues, that it should be 'Dave', only listened for generalities, and the BOT passed, largely on that one, change of volume spoken.
 
The following users thanked this post: MK14

Offline EEVblogTopic starter

  • Administrator
  • *****
  • Posts: 38039
  • Country: au
    • EEVblog
Re: DaveClone
« Reply #42 on: May 15, 2023, 06:25:38 am »
Meanwhile, Apple's female system voice in an Aussie accent is called Karen. I have no idea why that is? See [Settings > Accessibility > Spoken Content > System Voice ] if you're interested in hearing "winner winner chicken dinner" uttered by a generic Cath and Kim.

She's real, and the voice of Siri, Garmin, Navman, and TomTom
https://en.wikipedia.org/wiki/Karen_Jacobsen


 
The following users thanked this post: MK14, AndyBeez

Offline vk6zgo

  • Super Contributor
  • ***
  • Posts: 7653
  • Country: au
Re: DaveClone
« Reply #43 on: May 15, 2023, 07:56:16 am »
The UK is an accent soup. Not only can you tell from which city someone originates but also, from which side of that city. London I believe has at least 10 distinct accent groups? I will also add another London accent group which is a type of social media accent. This whiney accent afflicts 20 somethings who grew up watching the Carcrashians on UTube. It's common English mixed with west coast L.A. Makes the speaker sound double dim  :-\

On the subject of AI, PCs/Macs have system voices in many accents too. Test your localisation by asking the accented voice to speak various phrases. I found the Scottish accent on a Mac is highly adept at swearing - for some reason?!  :rant:Haggis!

Meanwhile, Apple's female system voice in an Aussie accent is called Karen. I have no idea why that is? See [Settings > Accessibility > Spoken Content > System Voice ] if you're interested in hearing "winner winner chicken dinner" uttered by a generic Cath and Kim.

Comparing computer accents might make an interesting segment on the AmpHour/etc  :wtf:

++ I'll add here the worse British English accent has to be the northern Irish or Ulster accent. Curiously, the southern Irish accent is one of the nicest. Weirdest is the impenitrable Newfoundland accent which is some mysterious blend of Gaelic, French and f-knows. As for the Kiwis, sorry to hear it's still legal to have six on a Sunday.

You might enjoy the following;

 
The following users thanked this post: MK14, AndyBeez

Offline AndyBeez

  • Frequent Contributor
  • **
  • Posts: 856
  • Country: nu
Re: DaveClone
« Reply #44 on: May 31, 2023, 09:43:10 am »
New to iOS 17, Apple Personal Voice.

Quote
iOS 17 Will Let You Create a Voice Sounding Like You in Just 15 Minutes

Personal Voice will be integrated with another new accessibility feature called Live Speech, which will let iPhone, iPad, and Mac users type what they want to say to have it be spoken out loud during phone calls, FaceTime calls, and in-person conversations.

https://www.macrumors.com/2023/05/17/ios-17-personal-voice-feature/

The future is nearly here. IOS17 roundup: https://www.macrumors.com/roundup/ios-17/
 
The following users thanked this post: MK14

Online tszaboo

  • Super Contributor
  • ***
  • Posts: 7584
  • Country: nl
  • Current job: ATEX product design
Re: DaveClone
« Reply #45 on: May 31, 2023, 11:18:33 am »
Honestly, I've tried these text to speech voice cloning softwares.
I've noticed that my voice was already changing if I was reading something, probably more than with the TTS.
I've tried one that worked with something like a 20 second sample, you imagine you would need hours of sample to train it. And it got the tone quite alright. I think when we get the frameworks to process hours of recordings, and learn speech differences/dialects then it will be very difficult to distinguish. Because the TTS was pronouncing some words quite differently than I'm, well because not a native speaker and such.
 
The following users thanked this post: MK14


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf