Just wondering how other people are dealing with brief outages. With a STT or LLM etc.
For example, I’ve been using Deepgram Flux as STT but would have times when data was sent to them, but they wouldn’t respond. Then it would stall and then reconnect and work or could go on for most of the call, resulting in a failed call.
So I switched providers to AssemblyAI as I got better latency in testing. But I had 40 calls today with the same issue. I can see data was sent to them, but their server never responded.
I’ve created a workaround already for Flux so that if this happens, the watchdog then goes back and responds. "Sorry, I didn’t get that. “Can you repeat that?” The caller then repeats, which gives it another chance.
The livekit failover feature assumes in an outage the provider will reply with a failure. But in these small outage situations this doesn’t work.
So I was just wondering what other people are doing in these situations and how they have tackled the same problem.
There has to be a neat method. I can give logs if people request them.
Thanks Troy