WEBVTT 00:00:00.000 --> 00:00:03.580 Okay, so what if I told you you could get unlimited, 00:00:03.940 --> 00:00:08.220 completely free AI voice cloning? Right. Running 00:00:08.220 --> 00:00:10.199 right there on your own computer, you'd probably 00:00:10.199 --> 00:00:12.039 think, okay, what's the catch? Yeah, is it some 00:00:12.039 --> 00:00:14.080 big subscription? Do I need like massive cloud 00:00:14.080 --> 00:00:16.100 server? Exactly. But while that's kind of the 00:00:16.100 --> 00:00:18.519 bombshell here, this tech is real and it runs 00:00:18.519 --> 00:00:22.519 entirely offline. Total privacy, zero cost, full 00:00:22.519 --> 00:00:24.820 control. It's like having your own voice cloning 00:00:24.820 --> 00:00:27.839 lab right there on your desktop or laptop. Welcome 00:00:27.839 --> 00:00:30.440 to the deep dive. Hey everyone. Our goal today 00:00:30.440 --> 00:00:33.310 really is to give you the essentials. We're pulling 00:00:33.310 --> 00:00:36.090 out the key info you need to actually use this 00:00:36.090 --> 00:00:38.630 local AI. Yeah, we'll cover how the tools work, 00:00:38.750 --> 00:00:41.590 the mechanics, but also, critically, the sort 00:00:41.590 --> 00:00:44.289 of non -negotiable rules, the ethics around voice 00:00:44.289 --> 00:00:46.090 replication. We're going to talk about how this 00:00:46.090 --> 00:00:49.549 changes content creation. Huge changes. Introduce 00:00:49.549 --> 00:00:51.350 you to the platform you need. It's called Pinocchio. 00:00:51.729 --> 00:00:53.990 And then get into the really cool stuff, the 00:00:53.990 --> 00:00:56.049 pro moves, those advanced settings that take 00:00:56.049 --> 00:00:59.210 your clone from good to, well... basically perfect 00:00:59.210 --> 00:01:02.609 let's dive in so let's start with why this even 00:01:02.609 --> 00:01:05.730 matters because it's not some you know expensive 00:01:05.730 --> 00:01:08.329 corporate thing anymore at all our sources are 00:01:08.329 --> 00:01:12.390 clear ai voice cloning it works it's accessible 00:01:12.390 --> 00:01:15.030 now and it's genuinely changing how creators 00:01:15.030 --> 00:01:17.890 work it's a massive creative unlock yeah sure 00:01:17.890 --> 00:01:19.950 you can make unique character voices that's fun 00:01:19.950 --> 00:01:23.370 right but the real power i think it's automating 00:01:23.370 --> 00:01:26.909 your own workflow clone your own voice once and 00:01:26.909 --> 00:01:29.599 boom You never have to record that same boring 00:01:29.599 --> 00:01:33.599 intro or outro or promo read ever again. Think 00:01:33.599 --> 00:01:35.680 about that. And the consistency part. That's 00:01:35.680 --> 00:01:38.680 huge. Imagine having like a library of courses, 00:01:38.739 --> 00:01:41.260 hundreds of videos. Yeah. And your voice sounds 00:01:41.260 --> 00:01:43.700 exactly the same. Same energy, same tone. Even 00:01:43.700 --> 00:01:45.500 if you recorded half of it when you had a cold. 00:01:45.659 --> 00:01:48.939 That is the secret weapon. So many big creators 00:01:48.939 --> 00:01:51.299 are already doing this, saving like literally 00:01:51.299 --> 00:01:53.500 hundreds of hours. What do you think is the biggest 00:01:53.500 --> 00:01:56.239 potential time saver here for creators then? 00:01:56.599 --> 00:01:58.719 Automating all that narration. Yeah. And keeping 00:01:58.719 --> 00:02:01.040 that perfect consistency across everything you 00:02:01.040 --> 00:02:03.659 make. Automating content narration, maintaining 00:02:03.659 --> 00:02:06.280 perfect consistency across your entire library. 00:02:07.019 --> 00:02:09.960 Got it. Which naturally brings us straight to 00:02:09.960 --> 00:02:12.139 the ethics. Okay. Because this is powerful stuff 00:02:12.139 --> 00:02:14.580 and it's free now. Great power, great responsibility, 00:02:14.800 --> 00:02:17.860 right? Exactly. Rule number one, absolutely non 00:02:17.860 --> 00:02:22.020 -negotiable. Yeah. Consent -based audio only. 00:02:22.099 --> 00:02:25.169 Meaning. Meaning your own voice. Yeah. Or. Audio 00:02:25.169 --> 00:02:28.409 from someone who gave you explicit, rock -solid, 00:02:28.409 --> 00:02:32.750 verifiable permission to clone their voice. Period. 00:02:33.050 --> 00:02:35.650 And the flip side is just as important. Never 00:02:35.650 --> 00:02:38.349 use this for fraud, for deception, for harassing 00:02:38.349 --> 00:02:41.020 people. We're misinformation. Deepfakes. That's 00:02:41.020 --> 00:02:42.780 the dark side. Absolutely. You've got to follow 00:02:42.780 --> 00:02:44.919 the law, follow platform rules like YouTube's 00:02:44.919 --> 00:02:47.319 terms. The tech itself is neutral, right? Totally 00:02:47.319 --> 00:02:49.599 neutral. It's how we use it. The moment you use 00:02:49.599 --> 00:02:52.139 it to trick someone, that's when it gets dangerous. 00:02:52.979 --> 00:02:55.319 Yeah, and honestly, I still kind of wrestle with 00:02:55.319 --> 00:02:58.099 prompt drift myself sometimes, even just working 00:02:58.099 --> 00:03:00.699 with my own voice clone and figuring out that 00:03:00.699 --> 00:03:04.120 line, you know, between useful automation and 00:03:04.120 --> 00:03:07.060 something that feels kind of deep fakie. There's 00:03:07.060 --> 00:03:09.340 constant attention, real vigilant. Okay, wait, 00:03:09.400 --> 00:03:11.770 back up. Prompt drift. What's that exactly for 00:03:11.770 --> 00:03:13.750 folks who haven't heard the term? Oh, right. 00:03:13.870 --> 00:03:17.530 So prompt drift is when the AI, as it's generating 00:03:17.530 --> 00:03:21.490 really long chunks of speech from text, it starts 00:03:21.490 --> 00:03:24.210 to kind of lose the original voice's unique flavor. 00:03:24.469 --> 00:03:26.770 Oh, okay. Yeah, it might sound a bit flatter, 00:03:26.889 --> 00:03:29.189 maybe less emotional, slightly robotic even. 00:03:29.310 --> 00:03:31.289 It means you've got to step back, maybe treat 00:03:31.289 --> 00:03:33.310 the input text or the settings. But choosing 00:03:33.310 --> 00:03:35.710 to run this locally on your own machine, that 00:03:35.710 --> 00:03:38.139 actually helps with some risks, doesn't it? compared 00:03:38.139 --> 00:03:40.219 to cloud services. Oh, massively. That's the 00:03:40.219 --> 00:03:42.699 core benefit, really. Why go local? Total privacy. 00:03:42.900 --> 00:03:46.000 Your voice data, which is biometric data. Right. 00:03:46.219 --> 00:03:48.759 Sensitive stuff. Very sensitive. It never leaves 00:03:48.759 --> 00:03:50.460 your computer. Yeah. You're not uploading it 00:03:50.460 --> 00:03:52.460 somewhere where it could be hacked or sold or 00:03:52.460 --> 00:03:54.360 analyzed without your knowledge. And besides 00:03:54.360 --> 00:03:57.560 security, the cost is just zero. 100 % free, 00:03:57.719 --> 00:04:00.860 unlimited use. No tokens, no subscriptions. Once 00:04:00.860 --> 00:04:03.400 it's set up, it even works offline. You're in 00:04:03.400 --> 00:04:05.919 complete control. So what's the core defining 00:04:05.919 --> 00:04:09.219 benefit of choosing a local setup then? Complete 00:04:09.219 --> 00:04:12.159 privacy, total control, and zero ongoing cost 00:04:12.159 --> 00:04:14.659 for unlimited use. Simple. Okay, so how do we 00:04:14.659 --> 00:04:16.220 actually get this running? You mentioned a platform, 00:04:16.399 --> 00:04:18.759 Pinocchio. Yeah, Pinocchio. Think of it like 00:04:18.759 --> 00:04:22.839 Steam for gamers, or maybe an app store, but 00:04:22.839 --> 00:04:25.079 specifically for AI models. Okay. It's basically 00:04:25.079 --> 00:04:27.959 a simple way to install and run these complex 00:04:27.959 --> 00:04:31.699 AI tools on your own PC without needing to be 00:04:31.699 --> 00:04:33.980 a coding wizard. So installing Pinocchio itself 00:04:33.980 --> 00:04:36.639 is pretty standard, like any other app. Yeah, 00:04:36.660 --> 00:04:39.779 download, run the installer, easy stuff. But 00:04:39.779 --> 00:04:42.060 the real magic happens inside Pinocchio when 00:04:42.060 --> 00:04:44.310 you install the specific voice model. And the 00:04:44.310 --> 00:04:48.310 one recommended in the sources is E2F5 -TTS. 00:04:48.529 --> 00:04:51.629 That's one, E2F5 -TTS. It's a really good open 00:04:51.629 --> 00:04:53.689 source model. Yeah. Known for being powerful, 00:04:53.829 --> 00:04:56.550 but also not needing like a supercomputer to 00:04:56.550 --> 00:04:58.970 run. It's great for home setups, does emotion 00:04:58.970 --> 00:05:01.170 really well. Now, there's a critical rule here 00:05:01.170 --> 00:05:04.459 during the install. Inside Pinocchio. Oh, yeah. 00:05:04.500 --> 00:05:07.800 This is super important. When E2F5 TTS is installing 00:05:07.800 --> 00:05:09.879 and it does this automatically through Pinocchio, 00:05:09.980 --> 00:05:12.279 do not interrupt it. Let it run. Let it run completely. 00:05:12.439 --> 00:05:14.160 It's downloading code, dependencies, all sorts 00:05:14.160 --> 00:05:16.319 of stuff. If you stop it, even for a second. 00:05:16.399 --> 00:05:18.000 It breaks. It breaks. You'll get errors later. 00:05:18.139 --> 00:05:20.620 Just let it go until it clearly says 100 % done. 00:05:20.720 --> 00:05:24.670 Be patient. Why is that patience during the E2F5 00:05:24.670 --> 00:05:28.050 TTS install so vital? The model needs all its 00:05:28.050 --> 00:05:30.829 dependencies installed completely, without interruption, 00:05:30.990 --> 00:05:33.209 to avoid errors later. Makes sense. Okay, so 00:05:33.209 --> 00:05:36.810 installation's done. We open up E2F5 TTS. You 00:05:36.810 --> 00:05:38.990 called it the cloning cockpit. Yeah, kind of 00:05:38.990 --> 00:05:40.889 looks like one. You'll see a few modes. The sources 00:05:40.889 --> 00:05:43.689 are really clear here. Stick with basic TTS. 00:05:43.910 --> 00:05:46.589 Basic TTS. Why that one? Because it puts all 00:05:46.589 --> 00:05:48.990 the AI's power into making one voice sound as 00:05:48.990 --> 00:05:51.620 good as possible. Highest quality. Best fidelity. 00:05:51.839 --> 00:05:54.180 Okay. There is a multi -speech mode for conversations, 00:05:54.259 --> 00:05:57.699 but honestly, it tries to juggle multiple voices 00:05:57.699 --> 00:05:59.560 at once, and the quality for each individual 00:05:59.560 --> 00:06:02.319 voice takes a hit, a noticeable hit. So if you 00:06:02.319 --> 00:06:04.560 need two speakers, better to generate them separately 00:06:04.560 --> 00:06:07.819 in basic TTS. Exactly. Generate speaker A's lines, 00:06:08.019 --> 00:06:10.279 generate speaker B's lines, then just edit them 00:06:10.279 --> 00:06:12.100 together later in your audio or video editor. 00:06:12.240 --> 00:06:14.759 Much cleaner result. Got it. Now, you said the 00:06:14.759 --> 00:06:16.639 most critical part of all this is the reference 00:06:16.639 --> 00:06:19.019 audio. Absolutely fundamental. The sample of 00:06:19.019 --> 00:06:20.959 the voice you feed the AI. The sources hammer 00:06:20.959 --> 00:06:23.379 this point home. Clean input equals clean output. 00:06:23.740 --> 00:06:26.040 Garbage in, garbage out. So what are the best 00:06:26.040 --> 00:06:29.240 practices? Non -negotiable stuff. Okay, number 00:06:29.240 --> 00:06:34.540 one, environment. Record in a really quiet room. 00:06:35.209 --> 00:06:38.310 No fans, no air conditioning hum, no computer 00:06:38.310 --> 00:06:41.129 noise, no traffic outside, dead quiet. Number 00:06:41.129 --> 00:06:43.970 two, style. Speak naturally, like you're having 00:06:43.970 --> 00:06:46.129 a conversation. Don't try to sound like a radio 00:06:46.129 --> 00:06:48.350 announcer unless that's the voice you want cloned. 00:06:48.509 --> 00:06:51.149 Just be natural. And length. How much audio do 00:06:51.149 --> 00:06:54.149 we need? Minimum. About 10 to 15 seconds of clear 00:06:54.149 --> 00:06:56.430 speech. That's the baseline. More is generally 00:06:56.430 --> 00:06:58.810 better, maybe up to a minute or so. But 10, 15 00:06:58.810 --> 00:07:02.129 seconds is usually enough to start. And crucially, 00:07:02.230 --> 00:07:05.740 crucially important, that audio file. It must 00:07:05.740 --> 00:07:08.279 be only the voice. No background music, no sound 00:07:08.279 --> 00:07:09.980 effects, definitely no other voices in there. 00:07:10.040 --> 00:07:12.040 Just the clean voice you want to clone. I actually 00:07:12.040 --> 00:07:13.500 learned that the hard way. I spent ages trying 00:07:13.500 --> 00:07:15.779 to clone my voice. Couldn't figure out why it 00:07:15.779 --> 00:07:18.399 sounded weird. Turns out the super faint hum 00:07:18.399 --> 00:07:20.399 from my external hard drive, like way in the 00:07:20.399 --> 00:07:23.339 background, was messing it all up. These models 00:07:23.339 --> 00:07:26.899 are sensitive to noise. They really are. So what 00:07:26.899 --> 00:07:29.300 are the two most essential factors for clean 00:07:29.300 --> 00:07:32.519 reference audio? Record 10 -15 seconds of speech 00:07:32.519 --> 00:07:36.759 in a completely quiet room. Just the voice. Mid 00:07:36.759 --> 00:07:39.139 -roll sponsor, Reed Placeholder. Okay, let's 00:07:39.139 --> 00:07:40.720 get into the good stuff. The advanced settings. 00:07:40.879 --> 00:07:43.500 This is where you go from, you know, pretty good 00:07:43.500 --> 00:07:45.879 audio to, wow, that sounds flawless. This is 00:07:45.879 --> 00:07:48.420 the pro level. This is the pro level. And honestly, 00:07:48.579 --> 00:07:51.160 most people skip this, but they shouldn't. First 00:07:51.160 --> 00:07:54.879 easy win, text preparation. Just typing the script 00:07:54.879 --> 00:07:57.740 carefully. More than that. Punctuation. The AI 00:07:57.740 --> 00:08:00.399 actually reads the punctuation to figure out 00:08:00.399 --> 00:08:03.449 tone and pacing. Oh, interesting. Yeah. Quick 00:08:03.449 --> 00:08:05.370 question mark. Makes the voice go up at the end. 00:08:05.490 --> 00:08:08.470 Exclamation point. Adds energy. Commas. Add pauses. 00:08:08.509 --> 00:08:11.189 If you just type a block of text with no punctuation. 00:08:11.350 --> 00:08:14.110 It sounds flat. Robotic. Exactly. So punctuation 00:08:14.110 --> 00:08:16.709 matters. A lot. Okay. What else is in advanced 00:08:16.709 --> 00:08:19.550 settings? Real game changer. Seed control. This 00:08:19.550 --> 00:08:21.870 is about consistency. Seed control. Like a random 00:08:21.870 --> 00:08:24.550 number. Kind of. So the AI is great at matching 00:08:24.550 --> 00:08:27.329 your voice's sound. The timbre. Yeah. But the 00:08:27.329 --> 00:08:30.180 emotion. The inflection. That can vary each time 00:08:30.180 --> 00:08:32.159 you generate audio, even with the same text. 00:08:32.419 --> 00:08:35.220 The seed is the key to locking that down. Wait, 00:08:35.299 --> 00:08:37.960 whoa. Okay, so you find a generation that sounds 00:08:37.960 --> 00:08:41.779 perfect. The exact right emotional tone. And 00:08:41.779 --> 00:08:43.919 you can use the seed number to make it sound 00:08:43.919 --> 00:08:45.960 exactly like that every single time. Imagine 00:08:45.960 --> 00:08:48.799 scaling that perfect tone across like a thousand 00:08:48.799 --> 00:08:51.000 hours of course content. That's the power. It's 00:08:51.000 --> 00:08:53.960 huge. So when you get that perfect take, you 00:08:53.960 --> 00:08:57.600 need to find the seed number. the AI used. It's 00:08:57.600 --> 00:09:00.059 usually in a log file or some metadata associated 00:09:00.059 --> 00:09:02.059 with the generated audio file. Find the magic 00:09:02.059 --> 00:09:04.500 number. Find the magic number. Then back in the 00:09:04.500 --> 00:09:06.460 settings, there's usually a checkbox, like use 00:09:06.460 --> 00:09:08.879 random seed. You uncheck that. Right. And then 00:09:08.879 --> 00:09:10.879 you type your magic seed number into the seed 00:09:10.879 --> 00:09:14.320 field. From then on, using that seed, the AI 00:09:14.320 --> 00:09:17.000 will reproduce that exact same emotional delivery 00:09:17.000 --> 00:09:20.159 every time. Consistency solved. That's incredible. 00:09:20.340 --> 00:09:22.179 Okay, what else is useful in advanced? Definitely 00:09:22.179 --> 00:09:25.019 check the box for remove silences. This automatically 00:09:25.019 --> 00:09:27.259 trims out little awkward pauses between words 00:09:27.259 --> 00:09:29.679 or sentences. Makes it sound tighter, punchier. 00:09:29.840 --> 00:09:31.960 Much punchier. Essential for stuff like social 00:09:31.960 --> 00:09:34.519 media clips, TikToks, shorts. Keeps the energy 00:09:34.519 --> 00:09:37.580 up. Good tip. Anything else? Yeah, speed adjustment. 00:09:38.019 --> 00:09:41.870 Sometimes the AI's default pace. can feel just 00:09:41.870 --> 00:09:44.889 a tiny bit slow, a little unnatural. Tweaking 00:09:44.889 --> 00:09:47.549 the speed just slightly, like maybe 1 .05 or 00:09:47.549 --> 00:09:50.230 1 .1, can make it sound much more human, more 00:09:50.230 --> 00:09:52.049 conversational. So it's about experimenting. 00:09:52.250 --> 00:09:55.990 Listen, tweak, regenerate. Exactly. It's an iterative 00:09:55.990 --> 00:09:58.889 process. Listen, evaluate, adjust one setting, 00:09:58.990 --> 00:10:02.539 generate again. Until it's perfect. So if my 00:10:02.539 --> 00:10:04.799 tone keeps changing between audio files, what 00:10:04.799 --> 00:10:06.899 should I check first? Save and reuse the seed 00:10:06.899 --> 00:10:08.980 number in the advanced settings. That locks the 00:10:08.980 --> 00:10:11.840 tone. Got it. Seed number for consistency. Now 00:10:11.840 --> 00:10:13.820 let's quickly talk applications and maybe some 00:10:13.820 --> 00:10:15.940 troubleshooting. All right. Real world use. We 00:10:15.940 --> 00:10:18.539 mentioned multi -speech mode isn't ideal for 00:10:18.539 --> 00:10:21.059 quality. Basic TTS and editing is better. Okay. 00:10:21.220 --> 00:10:23.940 But think bigger picture. E -learning courses. 00:10:24.059 --> 00:10:26.200 You could generate 10, 20 hours of narration. 00:10:26.440 --> 00:10:29.700 High quality. No vocal strain for you. Wow. Yeah. 00:10:29.799 --> 00:10:32.879 Or audiobooks. Turning old blog posts into audiobooks. 00:10:32.899 --> 00:10:35.220 Totally. In your own voice. Or turning out tons 00:10:35.220 --> 00:10:37.539 of short voiceovers for social media ads or updates. 00:10:37.700 --> 00:10:41.519 Fast. It really shifts content creation from 00:10:41.519 --> 00:10:45.600 this manual grind to something more automated, 00:10:45.820 --> 00:10:49.200 more scalable. Exactly. Now, quick troubleshooting. 00:10:49.220 --> 00:10:52.000 If the voice sounds robotic. Check the reference 00:10:52.000 --> 00:10:54.419 audio. Record it again in a quieter room. Perfect. 00:10:54.580 --> 00:10:56.899 If there are too many weird pauses. Go find that 00:10:56.899 --> 00:10:59.299 remove silences checkbox in the advanced settings. 00:10:59.519 --> 00:11:02.080 Yep. And installation errors. If things just 00:11:02.080 --> 00:11:04.340 aren't working after install. Delete the model, 00:11:04.580 --> 00:11:07.379 reinstall it carefully, and absolutely let it 00:11:07.379 --> 00:11:10.039 finish 100 % without interruptions. I got it. 00:11:10.059 --> 00:11:12.059 That covers like 90 % of the common problems. 00:11:12.379 --> 00:11:14.919 Besides saving time on, you know, daily little 00:11:14.919 --> 00:11:17.659 voiceover tasks, what's the biggest long -term 00:11:17.659 --> 00:11:20.399 productivity gain here? Producing massive amounts 00:11:20.399 --> 00:11:22.470 of hype. quality course or audiobook content 00:11:22.470 --> 00:11:24.809 really really quickly producing massive amounts 00:11:24.809 --> 00:11:27.370 of high quality course or audiobook content quickly 00:11:27.370 --> 00:11:31.269 opens up new possibilities okay so wrapping up 00:11:31.269 --> 00:11:34.350 we've done a deep dive into setting up free private 00:11:34.350 --> 00:11:37.250 unlimited voice cloning using Pinocchio using 00:11:37.250 --> 00:11:41.590 that e2f5 PTS model and the keys to really making 00:11:41.590 --> 00:11:44.909 it work seem to be number one super clean reference 00:11:44.909 --> 00:11:48.009 audio And number two, getting comfortable with 00:11:48.009 --> 00:11:50.389 those advanced settings, especially using seed 00:11:50.389 --> 00:11:53.269 control for consistency. Nail those two and you're 00:11:53.269 --> 00:11:56.389 golden. Yeah. And remember the ethics. The tools 00:11:56.389 --> 00:11:59.230 are neutral. How you use them isn't. Right. Use 00:11:59.230 --> 00:12:01.950 it to enhance your own stuff. Save yourself time. 00:12:02.049 --> 00:12:04.929 Make your work more accessible. Great. But avoid 00:12:04.929 --> 00:12:07.470 anything that involves impersonation, fraud, 00:12:07.690 --> 00:12:10.789 deception. Don't use it to trick people. Elevate. 00:12:10.789 --> 00:12:14.669 Don't erode trust. Exactly. Use this power responsibly. 00:12:15.179 --> 00:12:17.240 So here's a final thought to leave you with. 00:12:17.320 --> 00:12:19.720 This tech, this really powerful voice cloning, 00:12:19.879 --> 00:12:23.000 it's now free. It runs locally. Anyone with a 00:12:23.000 --> 00:12:24.779 decent computer can use it. Yeah. The barrier 00:12:24.779 --> 00:12:28.860 to entry just vanished. So what does that mean 00:12:28.860 --> 00:12:30.919 for the future? What kind of checks and balances 00:12:30.919 --> 00:12:33.740 will big platforms, you know, YouTube, Spotify, 00:12:34.039 --> 00:12:36.779 social media, what will they need to build internally 00:12:36.779 --> 00:12:39.460 to make sure content is authentic to prevent 00:12:39.460 --> 00:12:42.259 misuse on a massive scale? That's the big question 00:12:42.259 --> 00:12:44.070 now, isn't it? Something to definitely think 00:12:44.070 --> 00:12:46.110 about as you start playing with this. It is. 00:12:46.110 --> 00:12:47.889 We're excited to see what you create. Go build 00:12:47.889 --> 00:12:51.330 amazing things responsibly. Until next time.