TTS providers & voices

Supported runtime#

The current cloud adapter is Google Cloud Text-to-Speech. Provider-type labels for other vendors are configuration metadata; additional cloud synthesis adapters are not implemented.

Configure credentials#

Store a credential reference, not the JSON secret, in the provider record. Google resolution supports Application Default Credentials, file:// and env:// references. The API and Voice worker must be able to resolve the selected reference in their own service environments.

adc
file:///secure/location/google-service-account.json
env://GOOGLE_TTS_CREDENTIALSCopy

These are example formats. Keep actual credentials outside the website, Git and public demo.

Voices and synchronization#

Open TTS Voices, select the correct provider, and synchronize its catalogue after credentials work. Choose a voice ID and language matching the route's TTS language. A catalogue record does not prove that synthesis is authorized.

Audio formats#

Google can supply PCMA/PCMU directly. A G729 route requests LINEAR16 and converts it to raw 8 kHz G729 using a compatible ffmpeg encoder. Synthesis follows the codec negotiated in the SIP answer.

Verify#

A voice test makes an external synthesis request and may be billable by the provider. Do not run it merely to browse the demo. Check the error response, credentials status, selected provider/voice and VOICE_TTS_TIMEOUT when synthesis fails. Runtime audio caching is not implemented in this release.