https://retellai.com/?utm_source=discord&utm_medium=seo logo
Powered by
We ask the person to double confirm their name in ...
# ❓・ai-evy-support
a
@AI Evy We ask the person to double confirm their name in the beginning of the call, and then the next node after they confirm their name, we have a node with this prompt: "Say exactly: 'Great, nice to meet you, {{firstName}}! Now, before we find a date and time, here’s a quick heads up on what to expect during your consultation. We’ll do a full clinical eye exam with advanced testing and detailed eye mapping — this is the same thorough process we’ve refined since 1998 across more than 2 million procedures. I’ll explain more after we get you scheduled. Just give me one moment while I pull up your details.'" but for some reason when the ai agent speaks it, it sounds very robot/scripted. While other parts of the call it flows better and has more variety in tonality and speed. That prompt makes it sound very monotone. Are there any adjustments I can make to it to avoid that?
a
Thanks for tagging me @Alexander. I'll look through relevant sources and get back to you shortly.
### Answer The robotic/monotone delivery is likely caused by using "Say exactly:" which forces the LLM to output a rigid, scripted block of text that the TTS then reads verbatim. Here are some adjustments to improve naturalness: 1. Remove "Say exactly" and use flexible instructions instead Rather than forcing exact wording, give the agent the key points to convey and let it generate more natural phrasing. [(1) ]() For example:
Copy code
Greet the caller warmly using their first name ({{firstName}}) . Then explain what to expect during their consultation: a full clinical eye exam with advanced testing and eye mapping, refined since 1998 across over 2 million procedures. Let them know you'll explain more after scheduling, and ask them to wait a moment while you pull up their details.
2. Break the node into smaller pieces Long scripted blocks tend to sound monotone. You can split this into multiple nodes or shorter instructions to create more natural pauses and variation. 3. Adjust voice settings You can adjust the voice response speed in your voice model settings. If the agent is speaking too quickly through dense text, slowing it down may help with naturalness. [(2) ]() 4. Try a different voice model Some TTS voice models handle longer passages better than others. Testing different voice providers may yield more natural-sounding delivery. [(3) ]() 5. Enable speech normalization Turn on the 
normalize_for_speech
setting, which converts text to its spoken form before audio generation and can help with more natural delivery. The core issue is that "Say exactly" removes the LLM's ability to generate conversational variations, which is what makes other parts of your call sound more natural.
​ To learn more, see the sources I considered: [[1] Debug guide](https://docs.retellai.com/build/conversation-flow/debug-guide) [[2] TTS](https://discord.com/channels/1186424376300679209/1387743989490389052) [[3] My agent is pronouncing em-dashes in responses saying, "A-T" or "O-T" or "OT"](https://community.retellai.com/t/my-agent-is-pronouncing-em-dashes-in-responses-saying-a-t-or-o-t-or-ot/743) ​
Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.
a
@AI Evy well I don't want the LLM to generate conversational variations. I want it to say those exact words every time. It all comes down to the tonality. A human could very well say those words in that sequence in a natural way. How can I make the AI do it?
a
Thanks for the follow-up, @Alexander. I'll get back to you soon.
cues.
### Answer If you want to keep the exact wording but improve tonality, here are TTS-level adjustments you can make: 1. Adjust Voice Temperature Lower the 
voice_temperature
value (closer to 0) for more stable speech, or raise it (up to 2) for more variant, expressive delivery. [(1) ]() [(2) ]() 2. Adjust Voice Speed Use 
voice_speed
(range 0.5–2) to control pacing. Slowing down slightly can help the TTS deliver long passages more naturally. [(1) ]() [(2) ]() 3. Enable Dynamic Voice Speed Set 
enable_dynamic_voice_speed
to 
true
to allow automatic speed adjustment based on conversation context. [(1) ]() 4. Set Voice Emotion If using Cartesia or MiniMax TTS, you can set 
voice_emotion
to values like 
calm
, 
happy
, or 
sympathetic
to add emotional tone. 5. Add Punctuation for Natural Pauses Insert em-dashes, ellipses, or commas in your script to create natural pauses and breathing room. This can help break up monotone delivery without changing the words. 6. Try a Different Voice or Voice Model Some voices handle longer scripted passages better than others. Experiment with different 
voice_id
options or 
voice_model
settings (e.g., 
eleven_turbo_v2_5
, 
sonic-3
) to find one that delivers your exact text more naturally. Unfortunately, the TTS engine ultimately determines how scripted text sounds—there's no direct "tonality" or "emphasis" control beyond these settings. The monotone effect often comes from feeding the TTS a long, dense block of text without variation
​ To learn more, see the sources I considered: [[1] Create Voice Agent](https://docs.retellai.com/api-references/create-agent) [[2] Voice inconsistency - stuttering and filler sounds (aaaa, ehhh, hmmm)?](https://community.retellai.com/t/voice-inconsistency-stuttering-and-filler-sounds-aaaa-ehhh-hmmm/769) ​
Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.
a
@AI Evy these are the only settings I see in the node. Where is the Adjust voice temperature or is that not possible on node level? Node Settings Global Node Allow other nodes jump to this node without edges Node-level overrides Override agent defaults for this node Voice Speed Response Eagerness Interruption Sensitivity 0.00 blocks interruptionstops immediately LLM Choose a different LLM for t
a
Thanks for the follow-up, @Alexander. I'll get back to you soon.
### Answer Voice temperature is **not available as a node-level override**—it's an agent-level setting only. [(1) ]() Based on the documentation, the node-level overrides you're seeing (Voice Speed, Response Eagerness, Interruption Sensitivity, LLM) are the only ones available at the node level. [(2) ]() [(3) ]() To adjust 
voice_temperature
, you need to update it at the agent level using the Update Voice Agent API or in your agent's general settings. [(1) ]()
Copy code
javascript
import Retell from 'retell-sdk';
const client = new Retell({
  apiKey: 'YOUR_RETELL_API_KEY',
}) ;
const agentResponse = await client.agent.update('16b980523634a6dc504898cda492e939', {
  voice_temperature: 1.5,
}) ;
The 
voice_temperature
setting controls how stable the voice is, with values ranging from 0 to 2. Lower values produce more stable speech, while higher values create more variant speech generation. This setting currently only applies to 11labs voices. So for your use case, you'd need to set the voice temperature at the agent level, and it will apply across all nodes in your conversation flow.
Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.
​ To learn more, see the sources I considered: [[1] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[2] Update Conversation Flow](https://docs.retellai.com/api-references/update-conversation-flow) [[3] Create Conversation Flow](https://docs.retellai.com/api-references/create-conversation-flow) ​
a
@AI Evy what about the eagerness setting, what does that do? Examples based on setting level
a
Thanks for the follow-up, @Alexander. I'll get back to you soon.
### Answer --- I don't have specific documentation about the "Response Eagerness" (or "Eagerness") node-level setting in the sources available to me. The sources I reviewed cover node settings like Skip Response, Global Node, Block Interruptions, LLM override, and Fine-tuning Examples at the node level. [(1) ]() At the agent level, the documentation mentions Responsiveness, which controls how responsive the agent is—lower values mean the agent waits more and responds slower, with reducing responsiveness by 0.1 adding 0.5 seconds of agent wait time. [(2) ]() [(3) ]() However, I cannot confirm whether "Response Eagerness" at the node level functions the same way or has different behavior, as this specific setting isn't detailed in my available sources. Check out these potentially useful sources: - [Conversation Node]() - [Step 1: Configure global settings]() - [Update Conversation Flow]() Would you like me to try to answer generally based on my knowledge instead?
​ To learn more, see the sources I considered: [[1] Conversation Node](https://docs.retellai.com/build/conversation-flow/conversation-node) [[2] Step 1: Configure global settings](https://docs.retellai.com/build/conversation-flow/global-setting) [[3] Voice inconsistency - stuttering and filler sounds (aaaa, ehhh, hmmm)?](https://community.retellai.com/t/voice-inconsistency-stuttering-and-filler-sounds-aaaa-ehhh-hmmm/769) ​
Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.
a
@AI Evy what about the voice speed then
a
Thanks for the follow-up, @Alexander. I'll get back to you soon.
### Answer The 
voice_speed
setting controls the speed of the agent's voice, with values ranging from 0.5 to 2. [(1) ]() [(2) ]() [(3) ]() - Lower values (closer to 0.5) = slower speech - Higher values (closer to 2) = faster speech rate - **Default value**: 1 (if unset) Examples by setting level: - 
voice_speed: 0.5
— Slowest setting, noticeably slower delivery - 
voice_speed: 0.75
— Moderately slow - 
voice_speed: 1.0
— Default/normal pace - 
voice_speed: 1.5
— Faster than normal - 
voice_speed: 2.0
— Fastest setting You can also enable 
enable_dynamic_voice_speed
(set to 
true
) to allow automatic speed adjustment based on the user's speech rate and conversation context. For your scripted node that sounds monotone, try lowering the voice speed slightly (e.g., 0.8–0.9) to give the TTS more time to deliver the dense text, which may help with naturalness.
​ To learn more, see the sources I considered: [[1] Get Voice Agent](https://docs.retellai.com/api-references/get-agent) [[2] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[3] Create Voice Agent](https://docs.retellai.com/api-references/create-agent) ​
Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.
a
@AI Evy Is there any reliable way I can add more time gap between "... you scheduled" and "Just give me one moment while I pull up your details."" in the prompt here? "Great, nice to meet you, {{firstName}}! Now, before we find a date and time, here’s a quick heads up on what to expect during your consultation. We’ll do a full clinical eye exam with advanced testing and detailed eye mapping — this is the same thorough process we’ve refined since 1998 across more than 2 million procedures. I’ll explain more after we get you scheduled. Just give me one moment while I pull up your details."
a
Thanks for the follow-up, @Alexander. I'll get back to you soon.
### Answer --- Yes, you can add pauses using spaced dashes. The key is that spaces around the dash (
-
) are important for proper pausing behavior. [(1) ]() For a standard pause, use:
Copy code
I'll explain more after we get you scheduled. - Just give me one moment while I pull up your details.
For longer pauses, add multiple dashes with **double spaces between them**:
Copy code
I'll explain more after we get you scheduled. - - - - Just give me one moment while I pull up your details.
So your full prompt would look like:
Copy code
Great, nice to meet you, {{firstName}}! Now, before we find a date and time, here's a quick heads up on what to expect during your consultation. We'll do a full clinical eye exam with advanced testing and detailed eye mapping — this is the same thorough process we've refined since 1998 across more than 2 million procedures. I'll explain more after we get you scheduled. - - - - Just give me one moment while I pull up your details.
Important notes: - Some users have reported the dash method causing unexpected sounds with certain TTS providers and voices. [(2) ]() - If you're using ElevenLabs, some community members have had success with SSML tags like 
<break time="3s" />
wrapped in a 
<speak>
tag, though the official documentation focuses on the spaced dash method. [(3) ]() [(2) ]() If the dash method doesn't work reliably with your voice, you may need to experiment with different approaches or test with different voices.
​ To learn more, see the sources I considered: [[1] Add pause or read slowly](https://docs.retellai.com/build/add-pause) [[2] " - " for pausing is not working](https://discord.com/channels/1186424376300679209/1407867520492245023) [[3] Inserting a pause in the converation](https://community.retellai.com/t/inserting-a-pause-in-the-converation/268) ​
Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.
PreviousNext
Powered by