https://livekit.io logo
Join Slack
Powered by
Wanted to flag a strange issue we're noticing in t...
# ask-ai
b
Wanted to flag a strange issue we're noticing in the LiveKit voice agent pipeline — when the agent speaks, some completely unrelated background noise is audible in the audio. It's not static or echo — it sounds like external sounds such as laughter, music, or other ambient noise that has no relation to the agent or the call. This seems unusual and could indicate audio from another source bleeding into the agent's audio track. Please help us here.
✅ 1
t
Thanks for reaching out, Yethu. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
Thanks — I’m sorry you’re hearing unrelated background sounds during agent speech. I’m responding to your report about audio bleed in the LiveKit voice agent pipeline. Possible causes we can check quickly: • Another participant/agent is publishing audio into the same room and that track is being mixed into the agent output. • The agent isn’t restricted to a fixed participant (RoomIO) and is picking up multiple sources. • Application-level misrouting where received audio is being forwarded or published instead of only publishing TTS audio. • Use of RecorderIO/AudioStream or track-handling logic that inadvertently relays other tracks. Please provide the following so I can narrow this down: 1. Are you using a fixed participant identity for the agent (RoomIO fixed participant) or a dynamic/default setup? 2. Are multiple participants or other agents present in the same room when this happens? 3. Which SDK/language and LiveKit Agents library version are you using? 4. Do you have any custom audio forwarding, RecorderIO, or track-rebinding logic in your app? If yes, please paste the relevant snippet. If you prefer, I can also walk through steps to: • Verify participant tracks in Agent Console, • Enforce a fixed participant for the agent, • Audit publish/subscribe logic so only TTS is published. Tell me the answers above and I’ll provide targeted next steps. Sources: How to Enable Agent-to-Agent Speech Detection in LiveKit | How to set a fixed participant for agent audio processing | How telli automates high-volume enterprise phone operations with LiveKit an... | Text and transcriptions | LiveKit Documentation | Capturing the First Few Seconds of a Call for Processing, with interactive elements
b
eberuything is uniwue and only it i visible i the agent TTS
t
Thanks — I think your last message got garbled. I’m still investigating the audio-bleed into the agent TTS you reported. Could you clarify these quickly so I can narrow the cause? 1. Are you using a fixed participant identity for the agent (RoomIO) or a dynamic/default setup? 2. Are there any other participants or agents in the same room when this happens? 3. Which SDK/language and LiveKit Agents library version are you running? 4. Do you have any custom audio forwarding, RecorderIO, AudioStream, or track-rebinding code that might relay other audio? If so, please paste the short snippet. If you prefer, describe exactly what you meant by “eberuything is uniwue and only it i visible i the agent TTS.” , with interactive elements
b
everything is properly isolated and this is happening only in the agent TTS
t
Thanks — I appreciate you confirming isolation. I’m responding to your note that the noise appears only in the agent TTS output. Please answer these quick checks so I can pinpoint the cause: 1. Which TTS engine are you using (LiveKit built-in, AWS Polly, Azure, Google, custom)? 2. When you capture the raw TTS audio before publishing to LiveKit, does the unwanted background noise appear in that file? 3. Which SDK/language and LiveKit Agents library version are you running? 4. Do you apply any audio processing to TTS (mixing, effects, noise suppression, resampling) before publishing? If yes, paste the relevant snippet. 5. Are you publishing TTS as a 
LocalAudioTrack
(only) or are you also forwarding/relaying any received tracks with it? If you can attach a short sample of the raw TTS audio and the published agent output (or a spectrogram), that will help identify whether the noise is embedded in the TTS or introduced when publishing. , with interactive elements
b
I am not using these streams for my purpose apart form livekit using and I am using cartesia TS from inference
t
Thanks — and I appreciate you confirming isolation and that you’re using Cartesia TTS. To pinpoint whether the noise comes from Cartesia or from how you publish the audio to LiveKit, please provide the following: 1. Which LiveKit SDK/language and Agents library version are you using? 2. A short code snippet showing how you receive Cartesia TTS audio and publish it (creation of the 
LocalAudioTrack
/ publish call). 3. Does the background noise appear in the raw audio returned by Cartesia before you send it to LiveKit? 4. Are you doing any processing (resampling, mixing, normalization, noise suppression) to the TTS audio before publishing? If yes, paste the snippet. 5. Confirm you are publishing only the TTS 
LocalAudioTrack
(no forwarding/relay of other tracks). If you can attach a 5–10s sample of (a) the raw Cartesia TTS file and (b) the published agent output, that will let me tell you whether the noise is embedded in the TTS or introduced during publishing and provide exact next steps. , with interactive elements
2 Views
Open in Slack
PreviousNext
Powered by