Riva TTS Playback

AI & Data Science Deep Learning (Training & Inference) Riva
1

Please provide the following information when requesting support.

Hardware - GPU (A100/A30/T4/V100)
Hardware - CPU
Operating System
Riva Version NVIDIA_RIVA_SPEECH_SERVER_VERSION=25.05 | RIVA_SPEECH_VERSION=2.20.0
TLT Version (if relevant)
How to reproduce the issue ? (This is for errors. Please share the command and the detailed log here)

Riva doesn’t seem to support SSML tags such as time, telephone, digits. Using MagpieTTS Multilingual model is there a way to make the TTS playback of phone numbers, digits or time sound more natural? We tried sending phone numbers like 888-123-1234 or digits like 0-8-5-7-6 but still the playback doesn’t sound “normal”

3

Magpie TTS does not support SSML. The architecture has an LLM component so the model relies more on learned prosody patterns. Magpie supports the phoneme tag with custom pronunciation dictionary. Docs here. You can try normalizing the numbers before TTS like “eight eight eight, one two three” etc. to see if that helps with the naturalness.