✨(agents) add multilingual support for real-time subtitles

Add dynamic configuration for Deepgram STT via environment variables,
enabling multilingual real-time subtitles with automatic language
detection.

Changes:
- Add DEEPGRAM_STT_* environment variables pattern for configuration
- Implement _build_deepgram_stt_kwargs() to dynamically build STT
  parameters from environment variables
- Add whitelist of supported parameters (model, language) for LiveKit
  Deepgram plugin
- Log warnings for unsupported parameters (diarize, smart_format, etc)
- Set default configuration: model=nova-3, language=multi
- Document supported parameters in Helm values.yaml

Configuration:
- DEEPGRAM_STT_MODEL: Deepgram model (default: nova-3)
- DEEPGRAM_STT_LANGUAGE: Language or 'multi' for automatic detection
  of 10 languages (en, es, fr, de, hi, ru, pt, ja, it, nl)

Note: Advanced features like diarization and smart_format are not
supported by the LiveKit Deepgram plugin in streaming mode.

src/agents/multi-user-transcriber.py

@@ -29,15 +29,72 @@ logger = logging.getLogger("transcriber")
 
 TRANSCRIBER_AGENT_NAME = os.getenv("TRANSCRIBER_AGENT_NAME", "multi-user-transcriber")
 
+# Default Deepgram STT configuration
+DEEPGRAM_STT_DEFAULTS = {
+    "model": "nova-3",
+    "language": "multi",
+}
+
+# Supported parameters for LiveKit's deepgram.STT() in streaming mode
+# Note: Not all Deepgram API parameters are supported by the LiveKit plugin
+# detect_language is NOT supported for real-time streaming
+# Use language="multi" instead for automatic multilingual support
+DEEPGRAM_STT_SUPPORTED_PARAMS = {
+    "model",
+    "language",
+}
+
+
+def _build_deepgram_stt_kwargs():
+    """Build Deepgram STT kwargs from DEEPGRAM_STT_* environment variables.
+
+    Only parameters supported by LiveKit's deepgram.STT() are included.
+    Unsupported parameters are logged as warnings.
+    """
+    stt_kwargs = DEEPGRAM_STT_DEFAULTS.copy()
+
+    # Scan environment variables for DEEPGRAM_STT_* pattern
+    for key, value in os.environ.items():
+        if key.startswith("DEEPGRAM_STT_"):
+            # Extract parameter name and convert to lowercase
+            param_name = key.replace("DEEPGRAM_STT_", "", 1).lower()
+
+            # Check if parameter is supported by LiveKit plugin
+            if param_name not in DEEPGRAM_STT_SUPPORTED_PARAMS:
+                supported = ", ".join(sorted(DEEPGRAM_STT_SUPPORTED_PARAMS))
+                logger.warning(
+                    f"Ignoring unsupported Deepgram STT parameter: {param_name}. "
+                    f"Supported parameters: {supported}"
+                )
+                continue
+
+            # Parse value type
+            value_lower = value.lower()
+            if value_lower in ("true", "false"):
+                # Boolean values
+                stt_kwargs[param_name] = value_lower == "true"
+            elif value.isdigit():
+                # Integer values
+                stt_kwargs[param_name] = int(value)
+            else:
+                # String values
+                stt_kwargs[param_name] = value
+
+    logger.info(f"Deepgram STT configuration: {stt_kwargs}")
+    return stt_kwargs
+
 
 class Transcriber(Agent):
     """Create a transcription agent for a specific participant."""
 
     def __init__(self, *, participant_identity: str):
         """Init transcription agent."""
+        # Build STT configuration from environment variables
+        stt_kwargs = _build_deepgram_stt_kwargs()
+
         super().__init__(
             instructions="not-needed",
-            stt=deepgram.STT(),
+            stt=deepgram.STT(**stt_kwargs),
         )
         self.participant_identity = participant_identity
 

src/helm/meet/values.yaml

@@ -788,9 +788,21 @@ agents:
   ## @extra agents.envVars.FROM_CONFIGMAP.configMapKeyRef.key Key within a ConfigMap when configuring env vars from a ConfigMap
   ## @extra agents.envVars.FROM_SECRET.secretKeyRef.name Name of a Secret when configuring env vars from a Secret
   ## @extra agents.envVars.FROM_SECRET.secretKeyRef.key Key within a Secret when configuring env vars from a Secret
+  ## @extra agents.envVars.DEEPGRAM_STT_MODEL Deepgram model to use for speech-to-text (default: nova-3)
+  ## @extra agents.envVars.DEEPGRAM_STT_LANGUAGE Language code for transcription or 'multi' for automatic multilingual support with real-time code-switching (default: multi, supports: en, es, fr, de, hi, ru, pt, ja, it, nl)
   ## @skip agents.envVars
   envVars:
     <<: *commonEnvVars
+    # Deepgram Speech-to-Text configuration for real-time streaming
+    # Only 'model' and 'language' parameters are supported by the LiveKit plugin
+    #
+    # DEEPGRAM_STT_MODEL: "nova-3"              # Model selection (default)
+    # DEEPGRAM_STT_LANGUAGE: "multi"            # Multilingual mode with auto-detection (default)
+    # DEEPGRAM_STT_LANGUAGE: "fr"               # Force French only
+    #
+    # Note: Advanced features (diarization, smart_format, punctuate, detect_language)
+    # are NOT supported by the LiveKit Deepgram plugin in streaming mode.
+    # Use language="multi" for automatic multilingual support (10 languages).
 
   ## @param agents.podAnnotations Annotations to add to the agents Pod
   podAnnotations: {}