NVIDIA

NVIDIA Riva is a speech-to-text plugin that provides GPU-accelerated, low-latency speech recognition for the caller’s audio.

Setup

Set your NVIDIA API key in the worker environment. Generate a key from the NVIDIA build portal:

export NVIDIA_API_KEY=<key>

Import the plugin and pass it to the pipeline’s stt slot.

Python

from zrt.plugins import NvidiaSTT

stt = NvidiaSTT()

# Pipeline(stt=stt, ...)

Parameter	Type	Default	Description
`api_key`	`str \| None`	`None`	NVIDIA API key. Falls back to the `NVIDIA_API_KEY` environment variable when unset.
`model`	`str`	`"parakeet-1.1b-en-US-asr-streaming-silero-vad-sortformer"`	Riva/NIM ASR model used for transcription (Parakeet 1.1B with Silero VAD + Sortformer diarization).
`server`	`str`	`"grpc.nvcf.nvidia.com:443"`	Riva gRPC endpoint.
`function_id`	`str`	`""`	NVCF function ID for the hosted model.
`language_code`	`str`	`"en-US"`	BCP-47 language tag of the caller’s speech.
`sample_rate`	`int`	`16000`	Sample rate (Hz) of the input audio.
`use_ssl`	`bool`	`True`	Use a TLS/SSL gRPC channel.
`profanity_filter`	`bool`	`False`	Mask profanity in transcripts.
`automatic_punctuation`	`bool`	`True`	Add punctuation to transcripts.

Transcribed text passes to the LLM once turn detection decides the caller has finished speaking.