@millions-hair-24956 I send custom text transcription as

async def send_transcription(
        ctx: JobContext,
        participant: rtc.Participant,
        track_sid: str,
        segment_id: str,
        text: str,
        is_final: bool = True,
        language: str = "en",
):
    transcription = rtc.Transcription(
        participant_identity=participant.identity,
        track_sid=track_sid,
        segments=[
            rtc.TranscriptionSegment(
                id=segment_id,
                text=text,
                start_time=0,
                end_time=0,
                language=language,
                final=is_final,
            )
        ],
    )
    await ctx.room.local_participant.publish_transcription(transcription)

When language = "ja" (Japanese) or arabic or other language right-to-left written, my livekit FE client fails to differentiate transcriptions and it combines all agent replies. Works fine for other languages.

Sure thing! I'm looking through the available data to answer your question. Hold on a second... ⏳