https://retellai.com/?utm_source=discord&utm_medium=seo logo
Powered by
Hi ,we’re seeing significant audio quality degrada...
# ❓・ai-evy-support
d
Hi @AI Evy, we’re seeing significant audio quality degradation when using Retell over SIP compared to the Retell web preview for the exact same voice and content. We verified the SIP side thoroughly: SIP trunk: Telnyx Negotiated codec: Opus Sample rate: 48kHz SDP confirms: a=rtpmap:102 opus/48000/2 a=fmtp:102 useinbandfec=1; maxaveragebitrate=30000; ptime=20 Despite wideband Opus being negotiated, the audio over SIP sounds clearly worse than web preview: metallic artifacts on sibilants (s/sh/ch) reduced naturalness audible processing artifacts not present in preview This suggests something in the SIP audio pipeline may be altering the signal. Could you clarify: Do you apply any DSP on SIP output (noise suppression, normalization, compression, limiter, etc.)? Is there any internal resampling happening (e.g. 16kHz → 48kHz upsampling)? Is it possible to disable post-processing and receive raw TTS audio over SIP? Is SIP output expected to match web preview quality, or is degradation expected by design? Happy to provide call samples or full SIP traces if helpful.
a
Thanks for tagging me @Dziad. I'll look through relevant sources and get back to you shortly.
### Answer --- The available documentation does not provide specific details about: - Whether DSP (noise suppression, normalization, compression, limiter) is applied to SIP output - Internal resampling behavior (e.g., 16kHz → 48kHz upsampling) - Options to disable post-processing for raw TTS audio over SIP - Whether SIP output is expected to match web preview quality or if some degradation is by design What the documentation does confirm: - Different telephony providers have different audio codecs, and Retell's telephony integrations handle encoding/decoding internally—you don't need to manage it yourself [(1) ]() - For web calls, the frontend SDK abstracts away audio complexity, capturing PCM format audio - Retell's SIP server supports TCP (recommended) , UDP, and TLS transport methods, with SRTP media encryption available when using TLS [(2) ]() Since your specific questions about DSP processing, resampling, and quality parity between SIP and web preview are not addressed in the available sources, I'd recommend reaching out directly for clarification: - Visit [Contact Sales]() - Check [) Providing your call samples and SIP traces as you offered would likely help the team investigate the audio quality discrepancy. Would you like me to try providing a general answer based on common SIP/VoIP audio processing patterns instead?
​ To learn more, see the sources I considered: [[1] Audio Basics](https://docs.retellai.com/knowledge/audio-basics) [[2] Custom Telephony Overview](https://docs.retellai.com/deploy/custom-telephony) ​
Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.
Thanks for the follow-up, @Dziad. I'll get back to you soon.
d
@AI Evy Thanks for checking. To be clear: this is not a general question about SIP theory — we have a concrete reproducible issue. We verified on the wire that Opus 48kHz is negotiated (a=rtpmap:102 opus/48000/2), yet the same voice sounds significantly worse over Retell SIP than in Retell web preview. That strongly suggests the issue is inside Retell’s SIP audio pipeline, not in carrier configuration. Could someone from the engineering team please clarify: • whether any DSP is applied on SIP output • whether SIP output path differs from web preview path • whether raw / unprocessed audio over SIP is supported If needed, we can provide: • call recordings (web preview vs SIP) • full SIP traces • timestamps and Call IDs This is blocking production deployment on our side, so we’d really appreciate escalation to an engineer familiar with the SIP/media pipeline.
a
​ To learn more, see the sources I considered: [[1] Custom Telephony Overview](https://docs.retellai.com/deploy/custom-telephony) [[2] Poor audio quality for custom telephony dial to SIP endpoint](https://discord.com/channels/1186424376300679209/1279143924048466041) ​
### Answer I understand you have a concrete, reproducible audio quality issue with SIP compared to web preview—this is a legitimate concern. Unfortunately, the available documentation does not contain information about: - Whether DSP is applied on SIP output - Whether the SIP output path differs from the web preview path - Whether raw/unprocessed audio over SIP is supported What I can confirm from the documentation: The SIP server supports TCP (recommended) , UDP, and TLS transport methods, with SRTP media encryption available when using TLS. [(1) ]() Relevant community reports: Other users have reported similar audio quality issues with SIP. One user noted that when using the "dial to SIP endpoint" method, "the audio quality is very poor. With static and a lisp added to the speakers voice." A Retell team member suggested that "the dial to sip endpoint method would mean that the audio also needs to go through the telephony provider's programmable voice interface, thus might lead to lower quality" and recommended trying elastic SIP trunking instead for better audio quality. [(2) ]() Recommendation: Since you're using Telnyx SIP trunking (which is elastic SIP trunking) , and still experiencing degradation, this appears to require direct engineering investigation. Your specific questions about DSP processing, internal resampling, and audio pipeline differences are not addressed in the available sources. For escalation with your call recordings, SIP traces, and Call IDs: - Visit [Contact Sales]() - Check [)
Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.
Appreciate the feedback! Feel free to ask me again with additional guidance or context.
PreviousNext
Powered by