We’re developing a real-time translation system using Open AI’s streaming API. Initially, we chose a low-latency method where transcriptions are streamed and translated as they arrive. However, we’ve encountered a critical issue: when Speaker A speaks in longer or multi-part sentences, the translation output sometimes drops parts of the conversation, especially in the middle or end
Related topics
Topic | Replies | Views | Activity | |
---|---|---|---|---|
Transcribing live call (starting stream on Twilio call), after 1-4 min openAI starts to degrade | 0 | 68 | May 16, 2025 | |
Realtime API - Message being cutoff followed by silence | 1 | 392 | January 13, 2025 | |
Bad output when turn detection is not capturing complete thoughts | 0 | 238 | February 15, 2025 | |
The output audio does not fully match the output text; it ends early | 2 | 439 | October 11, 2024 | |
OpenAI_RealTime_Questions | 1 | 227 | February 20, 2025 |