Azure AI Speech has released Multilingual Post-Stream Refinement into public preview. This new feature removes the previous requirement of pre-selecting a single language for real-time transcription. It allows a single stream to automatically detect and transcribe multiple languages within the same session. This is a significant advancement as real-world speech often involves code-switching between languages.The refined transcript offers improved accuracy without increasing initial streaming latency. Internal testing shows an approximate 10% relative reduction in word error rate on average across Tier-1 locales. It provides even greater reductions for challenging content like long utterances and proper nouns. The system supports automatic language detection for 25 languages across 29 market locales.This public preview is available in six Azure regions. Users can leverage this feature with Speech SDK 1.50 or later and a speech resource in a supported region. Enabling it involves a simple configuration change to the SpeechConfig. This enhancement is particularly beneficial for applications that store or process final transcripts. Customers in various industries have already reported positive transcription quality gains. Feedback is encouraged as the feature moves toward general availability.