Voice
For the administrator. PULSE has no speech engine of its own. Everything it
says aloud goes to TIDE’s shared text-to-speech service, ttsd: the
voiceover on Reels and Stories PULSE makes, read-back in the app, podcast and
audiobook audio, and calls.
| Setting | Default | Effect |
|---|---|---|
TTSD_URL | http://ttsd.inference.svc.cluster.local:5005/v1/audio/speech (the in-cluster service) | The full URL PULSE posts text to. Set it to an empty value to turn voice output off. |
TTSD_CALLER_KEY_FILE | unset | The file holding PULSE’s service key for ttsd. It is read on every call, so a rotated key needs no restart. |
With voice output off, or ttsd unreachable, a Reel or Story PULSE makes is a
silent video that says “no voiceover: speech service unavailable”, and the
features that need speech say it is unavailable instead of returning silence.
Speech-to-text is separate: the voice tier at VOXTRAL_REMOTE_URL.
Making a video also needs ffmpeg on the server (the backend image has it;
FFMPEG_PATH overrides the binary). Without it, the app says “video is not
available on this server”.