WEBVTT 00:00.080 --> 00:00.480 Okay. 00:00.520 --> 00:02.760 So I'm going to go over to my workflow here. 00:02.760 --> 00:09.200 And I am going to go to the webhook and I'm going to take the test URL. 00:09.200 --> 00:10.440 I'm going to copy this. 00:10.760 --> 00:17.760 I'm now going to go over here, I'm going to paste in there that is the webhook URL. 00:17.960 --> 00:20.000 And now I'm going to start recording. 00:20.920 --> 00:21.960 Well hi there. 00:22.800 --> 00:24.160 Oh I have to say it loud while I use this site. 00:24.200 --> 00:24.640 Hang on. 00:25.000 --> 00:25.880 Well hi there. 00:25.920 --> 00:27.160 What is two plus two. 00:29.120 --> 00:29.880 Let's try this. 00:30.720 --> 00:31.600 Well hi there. 00:31.600 --> 00:32.800 What is two plus two? 00:33.600 --> 00:35.280 That seems to record quite nicely. 00:35.440 --> 00:37.960 And now I'm going to say send to N810. 00:37.960 --> 00:43.840 But first of course go back to N810 and go back to the beginning and say execute workflow. 00:43.840 --> 00:46.560 So it starts listening waiting for you to call the test. 00:46.840 --> 00:49.840 Go over here, send to N810 back over here. 00:51.040 --> 00:53.000 And we've got a problem right away. 00:53.000 --> 00:54.720 And it is a problem we're expecting. 00:54.920 --> 00:57.200 Cannot read properties of undefined. 00:57.240 --> 00:59.160 Let's double click and see what happens. 00:59.160 --> 01:04.480 So in came something called audio and the file that was looking for was called data. 01:04.520 --> 01:05.560 That's not right. 01:05.560 --> 01:07.840 So we have to change this to say audio. 01:08.000 --> 01:11.560 And now it knows that it's looking for a file that's called audio. 01:11.960 --> 01:12.480 Okay. 01:12.520 --> 01:13.400 Let's save it. 01:13.720 --> 01:14.520 Back we go. 01:14.760 --> 01:15.480 We go back in. 01:15.480 --> 01:18.720 Oh we have to press execute workflow back here. 01:18.720 --> 01:20.960 And now we can just press center N810 again. 01:22.040 --> 01:22.320 Back. 01:22.320 --> 01:22.960 Here we go. 01:23.440 --> 01:26.360 And now we've got a problem in node AI agent. 01:26.360 --> 01:27.640 It's going to be the same thing. 01:27.680 --> 01:29.440 All right so I double click here. 01:29.480 --> 01:30.400 Let's have a look. 01:30.680 --> 01:34.280 You see source it thinks connected chat trigger node. 01:34.280 --> 01:35.160 That's no good. 01:35.200 --> 01:37.720 We have to click here and say define below. 01:38.000 --> 01:41.560 Because as before we are going to give it some details of where to look. 01:41.560 --> 01:42.960 We're not going to tell it something fixed. 01:43.000 --> 01:45.320 We want it to evaluate an expression. 01:45.760 --> 01:48.240 And we're going to get the expression from over here. 01:48.240 --> 01:49.760 We want to take the text. 01:49.760 --> 01:50.360 Well hi there. 01:50.360 --> 01:51.520 What is two plus two? 01:51.680 --> 01:52.520 That's what we want. 01:52.560 --> 01:53.080 Isn't it cool. 01:53.080 --> 01:55.840 That is the transcribed text that came from 11 labs. 01:56.040 --> 01:58.440 Uh, so uh, it works. 01:58.440 --> 02:01.160 So we're going to take that and drop that in there. 02:01.160 --> 02:03.950 Or you could think in your mind, what do you think the expression is going to be? 02:04.550 --> 02:08.670 See, if you were right, it is simply JSON text as an expression. 02:08.670 --> 02:12.110 You can see the result that is now going to work. 02:12.430 --> 02:12.990 Okay. 02:13.310 --> 02:14.110 Escape back. 02:14.110 --> 02:19.470 We go to the canvas and now we're going to try it one more time okay. 02:19.590 --> 02:22.030 So we are going to press Execute workflow. 02:22.070 --> 02:23.390 We're going to go back here. 02:23.390 --> 02:26.750 We're going to press send to end and come back here off. 02:26.750 --> 02:27.390 It's going off. 02:27.390 --> 02:28.590 It goes off it goes. 02:28.630 --> 02:29.590 Two plus two is four. 02:30.150 --> 02:30.590 Ha. 02:31.710 --> 02:32.430 There we go. 02:32.550 --> 02:33.550 Worked first time. 02:33.750 --> 02:36.830 Uh and because that happened so fast I'm going to do it one more time. 02:36.830 --> 02:38.310 I wasn't expecting so fast. 02:38.510 --> 02:39.950 Uh, let's try this again. 02:39.950 --> 02:42.310 Let's, uh, press execute workflow. 02:42.910 --> 02:43.870 I'm going to go here. 02:44.510 --> 02:46.670 I'm going to go send to na n. 02:50.230 --> 02:50.670 Hello. 02:50.670 --> 02:51.790 Two plus two is way four. 02:52.110 --> 02:53.270 How can I help you today? 02:55.750 --> 02:56.910 Fabulous. 02:57.110 --> 02:58.230 So there you have it. 02:58.270 --> 03:01.470 We have a working audio app. 03:01.710 --> 03:06.030 It was a little bit of work, but it was great for you to see this as I promised. 03:06.070 --> 03:08.110 You see that Ann is in control. 03:08.110 --> 03:09.870 You've got the full workflow here. 03:09.870 --> 03:15.470 You've got a webhook, which is where the audio is posted in from our web page that gets transcribed 03:15.470 --> 03:18.830 with an API call and a text that goes through our AI agent. 03:18.830 --> 03:19.830 We're using Gemini. 03:19.870 --> 03:26.430 This time, Gemini converts the text back to speech again, and this node ensures that what gets replied 03:26.430 --> 03:34.070 to in this web request is in fact the audio file version of this, and that's why it replies with the 03:34.070 --> 03:34.710 audio. 03:34.830 --> 03:39.870 Congratulations, you've just made your first audio voice agent. 03:39.990 --> 03:41.790 I call it an audio voice agent. 03:41.790 --> 03:43.550 There are other kinds of voice agents. 03:43.550 --> 03:43.910 It is. 03:43.950 --> 03:45.430 It is a voice agent. 03:45.630 --> 03:49.550 Uh, and now, you know, the mission for you is to experiment with this. 03:49.590 --> 03:53.390 Obviously, I didn't put in a memory because we don't have, like, a chat history ID, so you'd have 03:53.390 --> 03:54.910 to futz around with that a bit. 03:54.910 --> 03:58.310 But do that, give it some memory, maybe even give it a tool or two. 03:58.350 --> 04:04.700 You're a pro with tools and then have a good old conversation Via this web page. 04:04.900 --> 04:10.420 And so the idea is that this web page that we've set up, this kind of fake web page could be anything. 04:10.420 --> 04:12.900 You could you could embed this in your WordPress site. 04:12.900 --> 04:14.780 You could turn this into something fancier. 04:14.820 --> 04:17.540 ChatGPT would have an easier time making this look prettier. 04:17.540 --> 04:22.140 And you can have it be something that will allow you to speak to your end flow. 04:22.140 --> 04:23.100 And you should. 04:23.100 --> 04:24.980 This is great experience. 04:24.980 --> 04:28.220 It's of course not the way I recommend we do it, which is what we're going to do next. 04:28.220 --> 04:31.380 But it's still a good way to see APIs in action. 04:31.380 --> 04:36.140 And if you had a bigger flow with lots of other things going on, but at some point you wanted to call 04:36.140 --> 04:38.380 out and have the audio side of it. 04:38.380 --> 04:42.620 This is exactly how you would do it as part of a bigger workflow. 04:43.460 --> 04:47.780 Okay, so that wraps up technique approach number one. 04:47.980 --> 04:49.100 And approach number two. 04:49.140 --> 04:51.740 As I said, it's more complicated in some ways. 04:51.740 --> 04:56.100 It's also going to be more simple, quicker for us to do perhaps because we've we've already got some 04:56.100 --> 04:57.180 experience with it. 04:57.180 --> 05:04.180 But let's, let's put put this one to bed and dive in to approach number two when 11 labs calls the 05:04.180 --> 05:04.860 shots.