Hi Everyone,
We're experiencing low audio quality using Hedra Realtime Avatar - need guidance:
Setup:
- livekit-plugins-hedra with Python Agents SDK
- TTS: Tries both ElevenLabs Turbo v2.5 and Azure TTS at 48kHz
- Audio output: 48kHz with RED encoding enabled
- Playback: Android WebView + macOS Safari (both exhibit issue)
Problem:
Audio sounds noisy, mechanical, distorted - despite excellent connection quality
Observed:
- Connection quality: EXCELLENT (both WebView and agent logs)
- Codec: audio/red (confirmed in trackSubscribed)
- Sample rate/bitrate: not exposed by WebRTC API
Investigation:
- Hedra plugin hardcodes SAMPLE_RATE = 16000 in DataStreamAudioOutput
- Direct TTS playback (no Hedra) sounds perfect through same WebView
- Suspect Hedra backend re-encodes at low Opus bitrate regardless of input
Questions:
1. Does Hedra API support audio quality parameters we're missing?
2. Is there a way to configure Opus bitrate for Hedra sessions?
3. Are there known Hedra backend limitations on audio quality?
4. Any recommended settings for production-quality audio with Hedra?
Thank you!