Mar 30 • General discussion 💬
Ghetto Smart Turn Detection FIX for Voice Agents
so ultravox is the only platform that has decent Smart Turn Detection model out of the box. But i was tired of voice agent interrupting the user on platforms like Retell/Vapi when user did not yet complete their sentence or still thinking about the response.
I tried putting this 1 sentence in system prompt of the voice agent and got amazing results out of it:
" - IF you think user is still thinking or DID NOT YET complete their sentence, just respond with "." OR "uh-huh", OR "ok" OR "yeah", based on the situation. this will make the conversation flow more natural and not interrupt the user when they are still thinking or have not completed their sentence."
if you tired of voice agent saying that its calling certain function (out loud to user) also try this:
"- DO NOT MENTION TOOLS OR FUNCTIONS IN YOUR RESPONSES to the user. they dont need to know that you are using a tool or function. if you must reply to user and do not have appropriate response, just respond with a dot : "."."
There both prompt snippets are working good both with gpt 4o and llama 3.3
I have attached the call recording (from retell) and also the transcript below:
User: Hello?
0:02
Agent: "Hi, may I speak with David Miller?"
0:04
User: Yes.
0:07
User: I'm him.
0:07
Agent: "Great! Hi David Miller, my name is Alex, and I'm a Recruiter with SwiftLogistics Inc. I'm following up on your interest in our Delivery Driver position. Is now still an okay time for a brief 10-15 minute chat?"
0:09
User: Yeah.
0:21
Agent: "Fantastic. Just to double-check, are you still actively exploring new opportunities and interested in the Delivery Driver role with us?"
0:23
User: Yes. Yes.
0:30
Agent: "Excellent. And generally speaking, what's your availability like to start a new role if things move forward?"
0:32
User: I'm available to start in
0:38
Agent: "Uh-huh."
0:44
User: two two weeks.
0:46
Agent: "Got it, thank you. This role requires frequent driving within the local area. Do you currently hold a valid driver's license?"
0:48
User: I do.
0:55
Agent: "Perfect. Could you tell me a bit about your previous experience with driving or delivery roles? What kind of work did you do?"
0:57
User: So recently,
1:04
Agent: "Okay."
1:09
User: I've been
1:11
Agent: "Uh-huh."
1:15
User: I've been, uh, work delivering groceries around
1:19
Agent: "Got it, delivering groceries. How long have you been doing that?"
1:27
User: I've been doing that for
1:32
Agent: "Yeah?"
1:37
User: Six
1:42
Agent: "Six months? Got it, thanks for sharing that. In delivery, sometimes things require a bit of extra effort. Can you tell me about a time you went above and beyond what was expected in a previous role, maybe to help a customer or solve a tricky delivery situation?"
1:46
User: Yeah. So there was a time when I
2:02
Agent: "Uh-huh."
2:08
User: was stuck in snow and
2:11
Agent: "Okay."
2:16
User: I had to park the car a few blocks away, and then
2:26
Agent: "Yeah?"
2:33
User: Walk to
2:37
Agent: "Uh-huh."
2:42
User: the customer and deliver groceries in their apartment And it was
2:44
Agent: "Wow, that sounds like a lot of effort. Thanks for sharing that example! It really shows your dedication. Now, things don't always go smoothly on the road. Could you describe a time you faced an unexpected challenge during a delivery – maybe significant traffic, a vehicle issue, or trouble finding a location – and how you handled it?"
2:53
User: Yes. There was a time when
3:11
Agent: "Okay."
3:16
User: I was
3:20
Agent: "Uh-huh."
3:26
User: stuck in traffic.
3:28
User: And then
3:30
Agent: "Yeah?"
3:34
User: I had to park the car in
3:38
Agent: "Okay."
3:46
User: few blocks away and then walk to the customer. And the customer was
3:49
Agent: "Uh-huh."
3:57
User: kinda sad that I was late, but still I explained, and then
3:59
Agent: "Got it, so you explained the situation to the customer and made sure they got their delivery. That shows good communication under pressure, thank you for sharing. Before we talk about next steps, do you have any quick questions for me about the role or SwiftLogistics at this stage?"
4:07
Agent: "Uh-huh."
4:31
User: I had that
4:32
Agent: "Yeah?"
4:36
User: How much do you guys
4:40
Agent: "That's a great question! Compensation and benefits details are best discussed with the Hiring Manager, Sarah Chen, in the next stage. Right now, I'd love to focus on learning more about your background and experience, if that's okay?"
4:47
User: I think that
5:02
Agent: "Uh-huh."
5:07
User: it's fine.
5:08
Agent: "Great. Based on our conversation, David Miller, I think you could be a strong fit, and I'd like to suggest the next step, which is a virtual interview with our Hiring Manager, Sarah Chen. Do you have your calendar handy to find a time?"
5:10
User: Calendar handy. I yes. I do have my calendar handy.
5:25
Agent: "Great. Sarah Chen is generally available Tuesday afternoon or Wednesday morning next week. What date, time, and importantly, time zone works best for you for a 30-minute video call?"
5:32
User: The best time for me is
5:44
Agent: "Yeah?"
5:49
User: Tomorrow afternoon,
5:53
Agent: "Got it. Just to clarify, tomorrow is Friday. What time and time zone works best for you in the afternoon?"
5:57
User: For me,
6:07
Agent: "Uh-huh."
6:11
User: Seven PM works good.
6:15
Agent: "Thanks for that. Just to confirm, 7 PM tomorrow in your time zone—could you let me know which time zone you're in?"
6:18
User: I'm in
6:25
Agent: "Okay."
6:30
User: eastern time zone.
4
0 comments
Ghetto Smart Turn Detection FIX for Voice Agents
skool.com/brendan
Learn To Make Money With AI!
- 50+ Free AI Agent Templates
- 60+ Free AI Course Videos
(n8n, Make, Vapi, Voiceflow)
- AI Software Discounts 💰
Leaderboard (30-day)
1
+77
2
+69
3
+60
4
+58
5
+57
Powered by