#Sometimes the voice agents sound distorted at the beginning of a call. They then stabilize after a
1 messages · Page 1 of 1 (latest)
Most early‑call “chipmunk/high‑pitched” artifacts on phone calls come from telephony codec/format mismatches. Configure your agent and integration to use telephony‑optimized formats end‑to‑end:
-
For Twilio calls, set BOTH:
- Agent TTS output format: μ-law 8000 Hz (Agent Settings → Voice)
- Agent input format: μ-law 8000 Hz (Agent Settings → Advanced)
ElevenLabs’ Twilio guides and troubleshooting explicitly call out matching Twilio’s μ-law 8 kHz requirement. (elevenlabs.io)
-
If you are in regions/carriers that expect A‑law, ElevenLabs added A‑law (8 kHz) support on April 7, 2025. Switch to A‑law 8000 Hz in your agent if that matches your telco path. (elevenlabs.io)
-
If you’re connecting via SIP trunking instead of Twilio, ensure your trunk negotiates G.711 (8 kHz) or G.722 (16 kHz), or that your system resamples correctly; ElevenLabs’ SIP deployment expects these codecs. (elevenlabs.io)
-
Prefer the native Twilio integration for inbound calls (the “custom server” guide is for outbound only); this reduces media/setup pitfalls. (elevenlabs.io)
-
If what you’re hearing is more like the voice briefly whispering/changing tone (not a codec pitch shift), that’s typically a voice/stability characteristic; trying a different voice or workflow (e.g., Studio for longer passages) can help. (help.elevenlabs.io)
Reference: ElevenLabs documents telephony‑optimized output/input formats (μ‑law/A‑law 8 kHz) in TTS and integration docs. Using non‑telephony formats (e.g., 44.1 kHz MP3) on phone lines can cause audio problems. (elevenlabs.io)
Sources:
Twilio outbound calls | ElevenLabs Documentation
April 7, 2025 | ElevenLabs Documentation
SIP trunking | ElevenLabs Documentation
Twilio custom server | ElevenLabs Documentation
Why does my voice start whispering, change accent, change tone, or break? – ElevenLabs
Text to Speech | ElevenLabs Documentation
I use the websocket connection!