The

transcript_confidence

value returned by the transcript for the Amazon Transcribe model is always 0 for every user message (while it has the normal expected range [0, 1] for other models, like Deepgram). Is there any known issue about this?

Thanks for reaching out, Luiz. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

Where can I find the return results from the Amazon Transcribe service?

Hey there! When you use the Amazon Transcribe service in LiveKit, the transcription results come through as

SpeechEvent

objects from the plugin. These events give you all the useful info like the transcribed text, start and end times, confidence scores, and even speaker identification if it’s available. Here’s how you can get to those results: • First, set up an event listener in your app to catch those transcription events. • The plugin sends out two main event types:

INTERIM_TRANSCRIPT

for partial, in-progress results, and

FINAL_TRANSCRIPT

for the completed transcriptions. • Each event includes handy details like

text

,

start_time

,

end_time

,

confidence

, and

speaker_id

. If you want a hand setting up the listener or working with these events, just give me a shout—I’m happy to help! Sources: AI voice agents | LiveKit Docs | uv.lock