We run a lot of calls for our clients but retell is making scaling difficult.
All of our calls recently (regardless of the TTS provider) have huge crackling making it near impossible to understand the agent. Latency on paper is average but on the phone its close to 3 seconds easily. It seems like something in the connection is breaking. This is further demonstrated by high LLM streaming failures which obviously add latency.
On top of that gpt5.6 is very smart and quick but sometimes it just does not generate a response at all. For example, if a user says ok to a question, it will just deem that not worth responding too. It is very strange behaviour from an LLm and i dont know if it is on your end or its just the model behaviour?
Hey @explitho Could you please create a new support ticket Support Help here and share the relevant Call IDs and a detailed description of the issue? This will help us investigate the problem more efficiently.
So the crackling is on every phone call regardless of the actual TTS provider - i have experienced it with platform voices, fish audio, elevenlabs, cartesia.
Here is a call ID call_db9943946cd152fcd785dfd202f
You can see the high rate of LLM streaming failures occuring. As well crackling which is not noticeable on the recording but happens on the actual phone call.
Hi @mark! I’d be happy to create a support ticket, but my account doesn’t have access to the Support Portal because it’s using a non-business email. Could someone either create the ticket for me, or let me know an alternative way to submit it?
I’d be happy to attach the audio file but I do not see an option to do so here. call_b932db6b9eb04d3ed8d891c62a3
Important to note that the recording is crystal clear without any issues, but on the actual call (which I recorded live on another device) the network quality is poor.
Anything on this? Crackling is absolutely killing my live calls this week, some people can barely hear anything. It is on every single call regardless of settings, TTS etc. It’s something to do with the retell pipeline. Maybe it is specific to my region. I have also tried stable and non stable server - no difference.
Hey @kruss6867 Could you please share the approximate timestamp in the call where the issue occurs? Also, let us know exactly what happens at that point. This will help us investigate the issue more efficiently.
I wish I could but it happens on the entire call on every single call. I can barely listen to it myself. It happens on my phone, my friends, my colleagues and the recipients.
To me, it sounds like some sort of unstable websocket connection. Hopefully this helps. It usually is ok for the first 10 seconds of the phone call and then gets bad after that and throughout the remainder of the call.
It looks like you’re using a cloned voice, which limits our ability to control the audio quality. We recommend switching to a standard voice or trying a cloned voice from a different provider that offers higher audio quality to see if the issue improves.
I have been using cloned voices for the past 18 months, plus this happens on every voice every TTS provider. Believe me it is not the voice. I have tried everything.
I don’t see a clear issue here. One thing I noticed is that the voice speed is very fast because you have increased the voice speed setting, and for the latency the prompt is 13k that will generate some latency
call_db9943946cd152fcd785dfd202f, v109
Please test a standard voice and share a call ID with the same issue if you believe the problem is not related to the cloned voice.
Here is one. Exact same problem platform standard voice. Nothing to do with the TTS engine. Latency from 13k tokens is marginally different from when i use a 4000 token prompt maybe 50-100ms difference so that is not the problem even the LLM actually shows that it is around 600ms but the e2e latency is sometimes up to 5000ms
Here is the voice problem call with the standard voice: call_a3b085e6ca755dfb5c1671099ef
Here is the long e2e latency even though the metrics of ttft are low: call_b4d92a48936d37a5427938c8b06 (see last turn of the call)
Anything new on this. This jitter is killing so many calls and people hanging up due to “bad lines” when its 100% the jitter, i know you will ask me for call ID’s but its on every voice, every TTS engine, every call. Not web calls. It has only started recently, it may look fine from a basic logs perspective but the websocket connection is either unstable or audio packets are being lost in the pipeline. I even tried switching the audio codecs on telnyx and switched them back nothing is working.
