diff --git a/api/services/configuration/registry.py b/api/services/configuration/registry.py
index c74b0116..aba0b7c7 100644
--- a/api/services/configuration/registry.py
+++ b/api/services/configuration/registry.py
@@ -584,6 +584,20 @@ class SarvamLLMConfiguration(BaseLLMConfiguration):
 
 
 OPENAI_REALTIME_MODELS = ["gpt-realtime-2"]
+# ISO 639-1 codes accepted by the Realtime API's input_audio_transcription.
+# Not exhaustive — the field allows custom input.
+OPENAI_REALTIME_LANGUAGES = [
+    "en",
+    "es",
+    "pt",
+    "fr",
+    "de",
+    "it",
+    "hi",
+    "ja",
+    "ko",
+    "zh",
+]
 OPENAI_REALTIME_VOICES = [
     "alloy",
     "ash",
@@ -618,6 +632,17 @@ class OpenAIRealtimeLLMConfiguration(BaseLLMConfiguration):
             "allow_custom_input": True,
         },
     )
+    language: str | None = Field(
+        default=None,
+        description=(
+            "ISO 639-1 language code for input audio transcription (e.g. 'pt', 'es'). "
+            "Improves transcription accuracy and latency. Leave unset to auto-detect."
+        ),
+        json_schema_extra={
+            "examples": OPENAI_REALTIME_LANGUAGES,
+            "allow_custom_input": True,
+        },
+    )
 
 
 GROK_REALTIME_MODELS = ["grok-voice-think-fast-1.0"]
diff --git a/api/services/pipecat/service_factory.py b/api/services/pipecat/service_factory.py
index e8a357a1..d34abd48 100644
--- a/api/services/pipecat/service_factory.py
+++ b/api/services/pipecat/service_factory.py
@@ -1008,6 +1008,13 @@ def create_realtime_llm_service(user_config, audio_config: "AudioConfig"):
             SessionProperties,
         )
 
+        # Pin the transcription language when configured. Without it the model
+        # auto-detects per utterance, which misfires on short/noisy telephony
+        # audio (e.g. Portuguese transcribed as English or Chinese).
+        transcription_kwargs = {}
+        if language:
+            transcription_kwargs["language"] = language
+
         return DograhOpenAIRealtimeLLMService(
             api_key=api_key,
             settings=DograhOpenAIRealtimeLLMService.Settings(
@@ -1015,7 +1022,9 @@ def create_realtime_llm_service(user_config, audio_config: "AudioConfig"):
                 session_properties=SessionProperties(
                     audio=AudioConfiguration(
                         input=AudioInput(
-                            transcription=InputAudioTranscription(),
+                            transcription=InputAudioTranscription(
+                                **transcription_kwargs
+                            ),
                         ),
                         output=AudioOutput(
                             voice=voice or "alloy",