### Answer Yes, you can configure the STT (Speech-to-Text) settings for an agent using the

stt_mode

parameter [(1) ]() [(2) ]() . The

stt_mode

parameter determines whether speech-to-text should focus on latency or accuracy . Available options are: -

fast

(default) -

accurate

-

custom

When you set

stt_mode

to

custom

, you must also provide the

custom_stt_config

object . Here's an example of how to update an agent with custom STT configuration:

javascript
import Retell from 'retell-sdk';
const client = new Retell({
 apiKey: process.env['RETELL_API_KEY'],
}) ;
const agentResponse = await client.agent.update('16b980523634a6dc504898cda492e939', {
 agent_name: 'Jarvis',
}) ;

[(1) ]() You can set these parameters when creating a new agent or updating an existing one [(1) ]() [(2) ]() .

​ To learn more, see the sources I considered: [[1] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[2] 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.

### Answer Based on the available documentation, I can confirm that STT configuration can be done through the API when creating or updating agents [(1) ]() . However, the sources don't explicitly describe whether these STT settings (

stt_mode

and

custom_stt_config

) are accessible through the dashboard UI. The global settings configuration in the conversation flow UI covers voice settings, language selection, LLM model, and speech settings like responsiveness and interruption sensitivity [(2) ]() , but the sources don't specifically mention STT mode configuration in that UI context. For the most reliable approach, you can configure STT settings using the API when updating your agent [(1) ]() . If you need to modify this through JSON, you can export your agent configuration from the dashboard (kebab menu next to the agent on the /agents page, then "Export") [(3) ]() , modify the

stt_mode

and

custom_stt_config

fields in the JSON, and reimport it . I'd recommend checking your dashboard UI directly to see if STT mode options are available in the agent settings, or using the API/JSON approach for guaranteed control over these parameters.

​ To learn more, see the sources I considered: [[1] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[2] Step 1: Configure global settings](https://docs.retellai.com/build/conversation-flow/global-setting) [[3] Discord message from @last2469 on 2026-03-18](https://discord.com/channels/1186424376300679209/1186462578310910112/1483765522033147905) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer To address both agent interruptions and long pauses simultaneously, you'll need to adjust two key parameters: responsiveness and interruption_sensitivity [(1) ]() . For your specific issue: Since you're experiencing both interruptions AND long pauses, this suggests the settings are conflicting. Here's the recommended approach based on a similar case: Decrease responsiveness - This controls how quickly the agent responds. A lower value (closer to 0) makes the agent wait more before responding, which can reduce interruptions [(1) ]() [(2) ]() . Increase interruption_sensitivity - This controls how easily users can interrupt the agent. A higher value (closer to 1) makes it easier for callers to interrupt, preventing the agent from talking over them . Key parameters to configure: - **responsiveness**: Value ranging from [0,1]. Lower value means less responsive agent (wait more, respond slower) , while higher value means faster exchanges. Default is 1 [(1) ]() . - **interruption_sensitivity**: Value ranging from [0,1]. Lower value means it takes longer/more words for user to interrupt agent, while higher value means it's easier for user to interrupt. Default is 1. When set to 0, agent would never be interrupted . Additional consideration for speakerphone calls: If your callers are using speakerphone, this problem can be more pronounced due to echo and reverb causing the system to misdetect end-of-speech [(2) ]() . The audio delay/latency on speakerphone might trigger premature turn-taking . You can also adjust the **reminder_trigge

r_ms** parameter, which triggers a reminder to the agent to speak if the user has been silent for the specified duration (default 10000 ms) [(1) ]() . Start by making small adjustments to these parameters and test the results with real calls to find the optimal balance for your use case.

​ To learn more, see the sources I considered: [[1] Create Voice Agent](https://docs.retellai.com/api-references/create-agent) [[2] Agent interrupting users on speakerphone calls](https://discord.com/channels/1186424376300679209/1442882951246512168) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer The cycle you're describing is a classic turn-taking conflict. Here's how to break it: Immediate adjustments: Increase responsiveness to 1.0 - Your current 0.8 setting is causing the agent to wait too long before responding, which creates the pause that prompts callers to repeat themselves [(1) ]() [(2) ]() . Setting responsiveness to 1 means faster exchanges and the agent responds when it can [(2) ]() . Decrease interruption_sensitivity to 0.5-0.6 - Your current 0.84 makes the agent stop too easily when the caller speaks [(1) ]() [(2) ]() . A lower value means it takes longer/more words for the user to interrupt the agent [(2) ]() , which prevents the agent from stopping prematurely when callers overlap. Why this works: Higher responsiveness reduces the pause that causes callers to think the agent didn't hear them . Lower interruption sensitivity prevents the agent from stopping every time there's slight overlap [(1) ]() [(2) ]() . This combination should reduce the repetitive cycle you're experiencing. Additional options to consider: You can add a general guideline in your prompt to "be patient and give time for the user to finish" [(3) ]() . This helps the LLM avoid drafting responses prematurely . Adjust reminder_trigger_ms (default 10000 ms) if you want to control how long the agent waits before reminding an unresp

onsive user [(1) ]() [(2) ]() . Test these settings with real calls and make small incremental adjustments from there to find the optimal balance for your specific use case.

​ To learn more, see the sources I considered: [[1] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[2] Create Voice Agent](https://docs.retellai.com/api-references/create-agent) [[3] Discord message from @theognis. on 2024-08-12](https://discord.com/channels/1186424376300679209/1186462578310910112/1272623893030637643) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer Boosted Keywords are custom vocabulary entries that bias the speech-to-text model toward specific words or terms to improve transcription accuracy [(1) ]() [(2) ]() . They're particularly useful for domain-specific terminology that the ASR model might not recognize correctly [(1) ]() . When to use Boosted Keywords: When specific words (like brand names, product names, or technical terms) are frequently misunderstood or missing from transcripts . For example, if users mention company names like "Figma" but it gets transcribed as "Sigma," or "Conga" gets transcribed as "Comdata" [(3) ]() . Key details: You can add up to 100 custom keywords [(1) ]() . These keywords instruct the transcriber model to have a bias for these words and pick them up in transcription [(4) ]() . They're configured in the agent's Speech Settings section in the dashboard [(2) ]() . Configuration: Boosted Keywords can be set using the

boosted_keywords

parameter when creating or updating an agent [(5) ]() [(6) ]() . This parameter accepts an array of strings representing the keywords you want to prioritize. This feature is especially helpful for improving recognition of names (company names, personal names) , industry-specific jargon, or any specialized vocabulary relevant to your use case [(1)

]() [(3) ]() .

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] Increase transcription accuracy](https://docs.retellai.com/reliability/wrong-transcript) [[2] Step 2: Configure the basic settings](https://docs.retellai.com/build/single-multi-prompt/configure-basic-settings) [[3] Discord message from @erika_jane9 on 2024-09-11](https://discord.com/channels/1186424376300679209/1186462578310910112/1283435022405079153) [[4] Speech Recognition Quality Issue](https://discord.com/channels/1186424376300679209/1224173869758414899) [[5] Update Voice Agent](https://docs.retellai.com/api-references/update-agent) [[6] Create Voice Agent](https://docs.retellai.com/api-references/create-agent) ​