ElevenLabs Scribe V2 STT not transcribing via LiveKit Inference (Deepgram works perfectly) Hi LiveKit team, I'm experiencing an issue where ElevenLabs Scribe V2 Realtime STT produces zero transcriptions via LiveKit Inference, while Deepgram Nova-3 works flawlessly with the exact same code.Environment • LiveKit Agents SDK (latest) • Railway deployment • Using

inference.STT()

object approach What Works (Deepgram)

from livekit.agents import AgentSession, inference

session = AgentSession(
    stt=inference.STT(model="deepgram/nova-3", language="en"),
    # ... other config
)
Result: stt_audio_duration=19.55, transcriptions appear correctly:DEBUG pipeline: user_input_transcribed - transcript='Hello? Can you hear me?' is_final=True language='en'
User said: Hello? Can you hear me?
STT metrics ✓

What Fails (ElevenLabs) - Tested ALL variations: Attempt 1: String shorthand

stt="elevenlabs/scribe_v2_realtime:en"

Result:

stt_audio_duration=0.0

, NO transcriptionsAttempt 2: inference.STT() with "en"

stt=inference.STT(model="elevenlabs/scribe_v2_realtime", language="en")

Result:

stt_audio_duration=0.0

, NO transcriptionsAttempt 3: inference.STT() with "en-US" (per docs)

stt=inference.STT(model="elevenlabs/scribe_v2_realtime", language="en-US")

Result:

stt_audio_duration=0.0

, NO transcriptionsLogs Comparison

Deepgram (works):Creating AgentSession with STT=inference.STT(model=deepgram/nova-3, language=en)
DEBUG track_flow: SUCCESS - Subscribed to AUDIO track
GAIN_BOOST: frame 2 AFTER +24.0dB - rmsdb=-38.9
DEBUG pipeline: user_input_transcribed - transcript='Hello? Can you hear me?' ✓
STT metrics ✓
stt_audio_duration=19.55 ✓
ElevenLabs (fails):Creating AgentSession with STT=inference.STT(model=elevenlabs/scribe_v2_realtime, language=en-US)
DEBUG track_flow: SUCCESS - Subscribed to AUDIO track
GAIN_BOOST: frame 2 AFTER +24.0dB - rmsdb=-30.5
[NO user_input_transcribed events]
[NO STT metrics events]
stt_audio_duration=0.0 ✗

Key Observations 1. Audio is being received (GAIN_BOOST logs show actual audio data with good signal: rmsdb=-27 to -30) 2. Model loads without errors (no "model not supported" message) 3. TTS and LLM work perfectly 4. Only STT fails with ElevenLabs 5. Switching ONLY the model name from

deepgram/nova-3

to

elevenlabs/scribe_v2_realtime

breaks transcription Questions 1. Is ElevenLabs Scribe V2 actually working in LiveKit Inference production? 2. Is there additional configuration required for ElevenLabs that's not in the docs? 3. Is this a known issue or bug? I need ElevenLabs with language which Deepgram Nova-3 doesn't support. Planning to use 5k+ minutes/month with LiveKit Inference add-on. Docs reference: https://docs.livekit.io/agents/models/stt/inference/elevenlabs/ Thanks for looking into this!

Hi, @limited-pillow-7054 I have replicated your issues, and I am going to feedback to the team about this For now can you use the plugin if possible while we are looking at this? session = AgentSession( stt=elevenlabs.STT( use_realtime=True, language_code="en", server_vad={ "vad_silence_threshold_secs": 0.8, "vad_threshold": 0.4, }, ), # A Large Language Model (LLM) is your agent's brain, processing user input and generating a response # See all available models at https://docs.livekit.io/agents/models/llm/ llm=inference.LLM(model="openai/gpt-4.1-mini"), # Text-to-speech (TTS) is your agent's voice, turning the LLM's text into speech that the user can hear # See all available models as well as voice selections at https://docs.livekit.io/agents/models/tts/ tts=inference.TTS( model="cartesia/sonic-3", voice="9626c31c-bec5-4cca-baa8-f8ba9e84c8bc" ), # VAD and turn detection are used to determine when the user is speaking and when the agent should respond # See more at https://docs.livekit.io/agents/build/turns turn_detection=MultilingualModel(), vad=ctx.proc.userdata["vad"], # allow the LLM to generate a response while waiting for the end of turn # See more at https://docs.livekit.io/agents/build/audio/#preemptive-generation preemptive_generation=True, )

@quaint-ambulance-32873 Great news, thanks. Was it api problem?

it was explained in the github issue that you have raised!