fix: Optimize response delay a bit

README.md

@@ -637,8 +637,8 @@ Conversation options are represented as features. They can be configured from Ap
 | `recognition_retry_max` | The maximum number of retries for voice recognition. | `int` | 2 |
 | `recording_enabled` | Whether call recording is enabled. | `bool` | false |
 | `slow_llm_for_chat` | Whether to use the slow LLM for chat. | `bool` | false |
-| `vad_cutoff_timeout_ms` | The cutoff timeout for voice activity detection in seconds. | `int` | 500 |
-| `vad_silence_timeout_ms` | The timeout for phone silence in seconds. | `int` | 500 |
+| `vad_cutoff_timeout_ms` | The cutoff timeout for voice activity detection in seconds. | `int` | 400 |
+| `vad_silence_timeout_ms` | The timeout for phone silence in seconds. | `int` | 400 |
 | `vad_threshold` | The threshold for voice activity detection. | `float` | 0.5 |
 
 ### Use an OpenAI compatible model for the LLM

app/helpers/features.py

@@ -29,11 +29,11 @@ async def callback_timeout_hour() -> int:
 
 
 async def vad_silence_timeout_ms() -> int:
-    return await _get(key="vad_silence_timeout_ms", type_res=int) or 500
+    return await _get(key="vad_silence_timeout_ms", type_res=int) or 400
 
 
 async def vad_cutoff_timeout_ms() -> int:
-    return await _get(key="vad_cutoff_timeout_ms", type_res=int) or 500
+    return await _get(key="vad_cutoff_timeout_ms", type_res=int) or 400
 
 
 async def vad_threshold() -> float:

cicd/bicep/app.bicep

@@ -905,8 +905,8 @@ resource configValues 'Microsoft.AppConfiguration/configurationStores/keyValues@
     recognition_retry_max: 2
     recording_enabled: false
     slow_llm_for_chat: false
-    vad_cutoff_timeout_ms: 500
-    vad_silence_timeout_ms: 500
+    vad_cutoff_timeout_ms: 400
+    vad_silence_timeout_ms: 400
     vad_threshold: '0.5'
   }): {
     parent: configStore