v2.6.39Latest
Speaking bots are heard when they speak, not three seconds later
- 1improved
- 1fixed
This release removes the three-second delay on a speaking bot's voice for good. v2.6.38 announced the fix, but the delay it removed was not the one holding the audio back; the remaining one is gone now, and a bot's reply reaches the meeting within tens of milliseconds of being produced. If you run speaking bots, this is the release that changes what your participants hear. There are no changes to API requests, responses or webhooks, and no action is needed.
Improvements
- Speaking bots are heard when they speak, not three seconds later. Audio sent to a bot through
streaming_config.input_urlused to be held for about three seconds before it reached the meeting, which added that much to every reply. v2.6.38 removed one buffer on this path, but measurements on real meetings showed the delay unchanged at 3.1 seconds because a second hold, at the start of the audio stream, was doing the same thing. That hold is gone. Measured the same way after this release, the bot's words land where its speech begins. See Streaming and Speaking bots.
Bug Fixes
- Transcripts attribute a speaking bot's words to the bot. The same delay put the last three seconds of every bot turn under the next human speaker's name and the human's last words before a reply under the bot's. With the audio arriving on time, speaker turns in the transcript line up with the recording. See Transcription.