Voice Demo for Jetson/L4T triton custom backends sources

I have been testing the Voice Demo for Jetson/L4T which I find really useful. However I need to access to the source of the triton custom backends in order to use it with models of different number of characters.
Particularly I would like to access to the sources of :

  • voice-actitivity-detector-trt-ctc-streaming/libvoice-activity-detector.so
    -ctc-decoder-cpu-trt-vad-streaming/libctcdecoder-cpu.so

Could it be possible? Otherwise how can I extend the asr_test to languages with different number of characters.

Hi @ivangonzaleztorre, those components aren’t open-source, they were an early sneak-peek of functionality from NVIDIA Jarvis. You may want to also check out NeMo which has been released.