:::: MENU ::::

The Switched ON Show

The Switched ON Show | Comedic Chaos and Stuff

Voice to Speech Results for SOS Episode 0

  • Comments Off on Voice to Speech Results for SOS Episode 0

Voice to Speech Results for SOS Episode 0

Latest Replies Forums The Orange Lounge Voice to Speech Results for SOS Episode 0

Viewing 8 posts - 1 through 8 (of 8 total)
  • Author
    Posts
  • #66083
    Ol’ Ben
    Participant

    I thought it would be fun to throw SOS episode zero into the ML Algorithm to see how it did detecting speech in the show, and well, it didn’t do too good. Below is what my computer and DeepSpeech think was said in show zero:

    i eenamost her near her measurements thessalonians erasistratus more than handsome attitudes coarseness characterises osiridean tanhaeuser aheer temperature theoretical verdaderamente teheran or another nursemaids but ochanee years irishwoman altogether has for the man aristagoras arachnoidea araucarias bearing up for charleston the sahuaros as sure otherwise careless shortages in trastevere i cried wauhatchie washers a sharp washerwoman anastasia realisation churchwardenship wished in the latournelles there rushed a house and a airships that aesahhahiyenenhon shotawhorora a fierceness he was warm melanochroi have harry and percolators persecution of the very emanations orientals were holocausted a coosh is wallerstaetten weatherwise idolisation alisander paraphrase a shasheeah or a leaseholder each and steerable bareheaded that mathematical or handshake chaeremonianus miyanoshita harewood rhadamanthine are unwearied hayowentha her afterall a salesperson anastasia mahometanism she would not secession rocketship idealised she is more arshtishena her lashes she tenderhearted these monotonous as sharers schatrenschar grateful he was aristocratic oversophisticated owaissa her average sharper hester harpocrates hartebeeste near shoesoles attenshun hildesheimer elaborate hypochondriachal as a charioteer honeythunder tollemache can you make position or haartebeestefontein upon rock rock at her anchorage nothing wherewithal ideas than the chatahoochie cachemere the panopolitan on chanticleer and armor he easterners awareness on each anathematised the artilleryman atalanta she roshinara to masticate mashalleed the championing a year hortense enchanted iterations onehorse her forrester had hardheartedness to her were easter a shareholder shasheeahs ierharerou ahaseragh yearnings are here he yelled teeter testamentaires the halicarnassian internationalisation walkover eisenstein a treatise in your car and you had no more moonshiners on a tacamahaca gerasim heart her apotheosis here and horsemanship than shiverarium attestation he anatomisation overestimate ashatea a charioteer so turnhaeuser shorter

    • This topic was modified 8 months, 1 week ago by Ol' Ben.
    #66085
    Ol’ Ben
    Participant

    Beware the dreaded chatahoochie cachermere….

    #66091
    Version3
    Keymaster

    Are any of those words we actually said? Are our accents that crazy?

    #66092
    Ol’ Ben
    Participant

    maybe “the” or “were”

     

    probably not “the halicarnassian internationalisation walkover”

    #66111
    Ol’ Ben
    Participant

    I know not many people use the forum, but I came back to this a few months later and I ran it using the “whisper.ai” models, and this is a sample of the first 60 seconds of show 000, the same as above.

    Detecting language using up to the first 30 seconds. Use --language to specify the language
    Detected language: English
    [00:00.000 –> 00:04.000] Okay, I need you to do something for me.
    [00:04.000 –> 00:06.000] Get up.
    [00:06.000 –> 00:09.000] And put your back on my shoulder.
    [00:13.000 –> 00:14.000] Oh gosh.
    [00:14.000 –> 00:15.000] I love that.
    [00:15.000 –> 00:16.000] Yeah?
    [00:16.000 –> 00:19.000] So, uh, this is a sample.
    [00:19.000 –> 00:21.000] And you’re not.
    [00:21.000 –> 00:23.000] Yeah.
    [00:23.000 –> 00:25.000] Yeah?
    [00:25.000 –> 00:27.000] Man, I gotta figure out how to do this thing, man.
    [00:27.000 –> 00:29.000] I don’t even see a button for it.

    [00:29.000 –> 00:39.000] The secret to setting up a new Apple idea is that you click on the create ID hyperlink.
    [00:39.000 –> 00:40.000] Oh, here it is.
    [00:40.000 –> 00:41.000] Genius.
    [00:41.000 –> 00:42.000] Off to the right.
    [00:42.000 –> 00:44.000] It’s amazing.
    [00:44.000 –> 00:47.000] There it is.
    [00:47.000 –> 00:51.000] Create my Apple id.
    [00:51.000 –> 00:54.000] I knew a guy named id once.
    [00:54.000 –> 00:57.000] Think I should do it for the United States or Turkey?
    [00:57.000 –> 00:58.000] Paraguay.

    So far, this is performing amazingly. Once I can tune the model to perform well on my machine, I will transcribe the first five episodes and post them here.

    There are a couple interesting imputations of the audio that it does, “put your back on my shoulder” and “and you’re not”. You can definitely hear something that sounds like that in the audio, which is amazing that it can pull that off.

    The timestamps indicate different speakers. It differentiated Bryan and Jerry perfectly in that 60 second clip.

    • This reply was modified 3 weeks, 5 days ago by Ol' Ben.
    #66113
    Ol’ Ben
    Participant

    This is the same 60 seconds run through the faster model.

    [00:00.000 –> 00:05.000] Okay, I need to do something.
    [00:05.000 –> 00:07.000] Get up.
    [00:07.000 –> 00:08.000] Get up.
    [00:08.000 –> 00:10.000] Get up.
    [00:10.000 –> 00:13.000] Get up.
    [00:13.000 –> 00:14.000] Oh gosh.
    [00:14.000 –> 00:15.000] I love that.
    [00:15.000 –> 00:16.000] Yeah.
    [00:16.000 –> 00:19.000] So, uh, this is a sample.
    [00:19.000 –> 00:21.000] You’re not.
    [00:21.000 –> 00:24.000] Yeah.
    [00:24.000 –> 00:25.000] Yeah.
    [00:25.000 –> 00:27.000] Man, I gotta figure out how to do this thing, man.
    [00:27.000 –> 00:29.000] Well, I don’t even see a button for it.
    [00:29.000 –> 00:39.000] The secret to setting up a new Apple idea is that you click on the, uh, create ID type.
    [00:39.000 –> 00:40.000] Oh, here it is.
    [00:40.000 –> 00:41.000] Genius.
    [00:41.000 –> 00:42.000] Off the right.
    [00:42.000 –> 00:45.000] It’s amazing.
    [00:45.000 –> 00:47.000] There it is.
    [00:47.000 –> 00:49.000] Create my Apple.
    [00:49.000 –> 00:51.000] Ed.
    [00:51.000 –> 00:53.000] I knew it got it.
    [00:53.000 –> 00:54.000] Months.
    [00:54.000 –> 00:57.000] Think I should do it for your United States or Turkey.

     

    You can see it misses intonation, more of the quiet parts are not picked out, and the prasing is much different.

     

    This run is through the small model that’s a tiny bit faster. The first line is hilarious:

    [00:00.000 –> 00:09.000] OK, I need you to do something for me, get up, and then I’m going to risk my life.
    [00:13.000 –> 00:15.000] Oh gosh, I love that.
    [00:15.000 –> 00:20.000] Yeah, so this is a sample and you’re not.
    [00:22.000 –> 00:23.000] Yeah.
    [00:23.000 –> 00:24.000] Yeah.
    [00:25.000 –> 00:27.000] Man, I gotta figure out how to do this thing, man.
    [00:27.000 –> 00:29.000] I don’t even see a button for it.
    [00:29.000 –> 00:37.000] The secret to setting up a new apple idea is that you click on the create ID hyper.
    [00:39.000 –> 00:40.000] Oh, here it is.
    [00:40.000 –> 00:41.000] Genius.
    [00:41.000 –> 00:42.000] Off the right.
    [00:42.000 –> 00:43.000] It’s amazing.
    [00:45.000 –> 00:46.000] There it is.
    [00:47.000 –> 00:49.000] Create my apple id.
    [00:51.000 –> 00:53.000] I knew it got an id once.
    [00:54.000 –> 00:57.000] You think I should do it for the United States or Turkey?
    [00:57.000 –> 00:58.000] Paraguay.

    • This reply was modified 3 weeks, 5 days ago by Ol' Ben.
    #66115
    Ol’ Ben
    Participant

    I ran the complete show 000 through the program and uploaded the results here. If you watch through VLC, you need to use a visualization to see the subtitles, Audio -> Visualizations. Or you can follow along in the text file.

    https://zoael.funnyvirus.cool/PUBLIC/SOS-Transcripts/

    I’d give it a solid 9.5/10 for accuracy. It’s great.

    I am going to run a later show with higher quality audio with the faster model to see how it does. I am doing show 41 and 143, good combination of voices, song game with garage band/keyboard, and a Corby phone call

     

    • This reply was modified 3 weeks, 5 days ago by Ol' Ben.
    #66117
    Ol’ Ben
    Participant

    Needless to say it did a great job of even picking up the song game with Jerry playing the piano in show 41:

Viewing 8 posts - 1 through 8 (of 8 total)
  • You must be logged in to reply to this topic.