Update game build from GitLab CI

server/ActionProcessor.py

@@ -43,18 +43,21 @@ class ActionProcessor(Thread):
 
             1. **Reconstruct the most logical full command** by analyzing the given fragments and arranging them into a coherent order.
             2. **Select the closest matching three-word action** from the following predefined options:
-                ["drop it", "don't drink", "None", "stay back", "stop touching", "move away"].
+                ["drop it", "don't drink", "None", "stay back", "stop touching", "move away", "none"].
                 Always choose the most contextually relevant option. If no match is appropriate, select "None."
-            3. **Determine the parenting sentiment**, which should always be classified as "negative."
+            3. Determine whether the parenting sentiment behind the command is positive or negative**, based on the following criteria:
+                - If the statement includes the baby’s name ("Harold"), it is considered **positive parenting**.
+                - If the statement consists only of direct commands without affectionate or encouraging words (e.g., "good boy," "better"), it is considered **negative parenting**.
 
-            Your response should strictly follow this JSON format:
+            Your response should strictly follow this JSON format as an array of strings:
 
-            {{
-                "RECONSTRUCTED_ORDER": "[reconstructed text]",
-                "ACTION": "[chosen action]",
-                "SENTIMENT": "negative"
-            }}
+            ```json
+            [action,sentiment]
+            ```
 
+            for example:
+            ["drop it", "negative"]
+            ["none", "positive"]
             Transcription fragments: "{input_text}"
             """,
             }
@@ -62,7 +65,16 @@ class ActionProcessor(Thread):
 
         response = self.mistral_client.chat.complete(
             model="mistral-large-latest",
-            messages=messages,
+            messages=messages
+            + [
+                {
+                    "role": "assistant",
+                    "content": "[",
+                    "prefix": True,
+                }
+            ],
+            response_format={"type": "json_object"},
+            temperature=0.0,
         )
 
         result: str = response.choices[0].message.content
@@ -77,12 +89,12 @@ class ActionProcessor(Thread):
         if len(self.text_buffers) < 3:
             return None
 
-        if len(self.text_buffers) >= 3:
+        if len(self.text_buffers) > 3:
             _ = self.text_buffers.pop(0)
 
-        candidate = self.text_buffers[-2]
+        candidate = self.text_buffers[1]
 
-        if len(self.text_buffers[0]) <= len(candidate) <= len(self.text_buffers[-1]):
+        if len(self.text_buffers[0]) < len(candidate) >= len(self.text_buffers[2]):
             sentiment_and_action = self.get_sentiment_and_action(candidate)
             return {"type": "action", "sentiment": sentiment_and_action}
 

server/AudioTranscriber.py

@@ -69,6 +69,8 @@ class AudioTranscriber(threading.Thread):
                         logger.info(
                             f"[{segment.start:.2f}s -> {segment.end:.2f}s] {segment.text}"
                         )
+                    else:
+                        self.action_queue.put("")
 
             except Empty:
                 continue  # If queue is empty, continue waiting