NVIDIA Riva Client

2.25.1 · active · verified Thu Apr 16

The `nvidia-riva-client` library provides Python client APIs for interacting with NVIDIA Riva speech AI services, including Automatic Speech Recognition (ASR), Text-to-Speech (TTS), and Neural Machine Translation (NMT). It enables developers to integrate advanced conversational AI capabilities into their applications. The current version is 2.25.1, with releases typically tied to major Riva platform updates, leading to a moderately frequent release cadence.

Common errors

```
grpc._channel._MultiThreadedRendezvous: <_MultiThreadedRendezvous of RPC that terminated with: status = StatusCode.UNAVAILABLE
```
cause The Riva server is not running, is inaccessible, or the `RIVA_URI` is incorrect/firewalled. This indicates a network or server availability issue.

fix Verify the Riva server is running and healthy. Check the `RIVA_URI` (e.g., `localhost:50051`, `your.server.ip:50051`) and ensure there are no firewall rules blocking the connection. If running in Docker, ensure ports are correctly mapped.
```
ModuleNotFoundError: No module named 'riva.client'
```
cause The `nvidia-riva-client` package is not installed or the Python environment is incorrect.

fix Install the package using `pip install nvidia-riva-client`. Ensure you are running your script in the same Python environment where the package was installed.
```
AttributeError: module 'riva.client' has no attribute 'NLPClient'
```
cause Attempting to use the NLP client, which has been removed in recent versions of the `nvidia-riva-client` library.

fix The NLP client was deprecated and removed in version 2.15.0 and later. If you need NLP capabilities, use `nvidia-nemo-client` or another dedicated NVIDIA NLP framework.
```
ImportError: cannot import name 'RecognitionConfig' from 'riva.client'
```
cause This error often occurs if there's a version mismatch between `nvidia-riva-client` and its `grpcio` dependency, or if the installed `nvidia-riva-client` is corrupted/incomplete.

fix Try reinstalling the client and its dependencies: `pip uninstall -y nvidia-riva-client grpcio && pip install nvidia-riva-client`. Ensure your Python version meets the `nvidia-riva-client` requirements (`>=3.7`).

Warnings

breaking The `grpcio` dependency often requires specific versions to match the `nvidia-riva-client` version. Incompatible `grpcio` versions can lead to `ImportError` or `AttributeError` during runtime, particularly with protobuf definitions or gRPC channel initialization.
Fix: Always install `nvidia-riva-client` directly via pip; it will pull the correct `grpcio` version. If issues persist, verify the `grpcio` version manually (`pip show grpcio`) against the `nvidia-riva-client` PyPI dependencies. Upgrade `nvidia-riva-client` to the latest version to get the widest compatibility.
breaking The NLP client APIs were deprecated and removed in Riva client versions starting from 2.15.0.
Fix: Migrate any NLP tasks (e.g., intent recognition, named entity recognition) to the separate `nvidia-nemo-client` library or alternative NVIDIA NLP frameworks. Do not attempt to use `riva.client.NLPClient`.
gotcha Riva client versions are generally designed to be compatible with a specific range of Riva server versions. Mismatched client and server versions can lead to unexpected errors, `gRPC UNAVAILABLE` status codes, or incorrect API behavior.
Fix: Always check the NVIDIA Riva documentation for the recommended client-server version compatibility matrix. It is generally best practice to keep the client and server versions as closely aligned as possible.
gotcha Authentication methods and metadata retrieval (e.g., `get_auth_metadata`) have seen updates across versions, which might break older authentication patterns.
Fix: Refer to the latest NVIDIA Riva client documentation and examples for the current recommended authentication practices, especially when dealing with secure or token-based server deployments. Use the `riva.client.Auth` class and its methods as demonstrated in up-to-date quickstarts.

Install

pip install nvidia-riva-client Install latest version

Imports

ASRClient
```
from riva.client import ASRClient
```
TTSClient
```
from riva.client import TTSClient
```
Auth
wrong:
```
from riva.client.auth import Auth
```
correct:
```
from riva.client import Auth
```
While a subdirectory exists, the official examples and common practice import Auth directly from `riva.client`.
AudioEncoding
```
from riva.client import AudioEncoding
```

RecognitionConfig

from riva.client import RecognitionConfig

Quickstart

This quickstart demonstrates how to set up the Riva client, connect to a Riva server, and perform a simple Automatic Speech Recognition (ASR) task on a local WAV file. Ensure the `RIVA_URI` environment variable is set to your Riva server's address (e.g., `localhost:50051`). The script creates a dummy audio file if one isn't present for demonstration.

import os
import riva.client
import time
import wave

# Configure Riva server connection
# Ensure RIVA_URI is set in your environment (e.g., 'localhost:50051' or a remote address)
# For authenticated connections, set RIVA_API_KEY if needed.
riva_uri = os.environ.get('RIVA_URI', 'localhost:50051')

# Simple WAV file for ASR (create a dummy one if not present)
dummy_audio_file = 'dummy_audio.wav'
if not os.path.exists(dummy_audio_file):
    with wave.open(dummy_audio_file, 'wb') as wf:
        wf.setnchannels(1)
        wf.setsampwidth(2)
        wf.setframerate(16000)
        wf.writeframes(b'\x00' * 16000 * 2) # 1 second of silence
    print(f"Created a dummy audio file: {dummy_audio_file}")

try:
    # Establish authentication (if needed, otherwise Auth() is sufficient)
    auth = riva.client.Auth(uri=riva_uri)

    # Initialize ASR client
    asr_client = riva.client.ASRClient(auth)

    # Configure ASR recognition
    config = riva.client.RecognitionConfig(
        encoding=riva.client.AudioEncoding.LINEAR_PCM, # or FLAC, MULAW, etc.
        sample_rate_hertz=16000,
        language_code="en-US",
        max_alternatives=1,
        enable_automatic_punctuation=True,
    )

    # Perform ASR on a local audio file
    print(f"Transcribing {dummy_audio_file} from Riva server at {riva_uri}...")
    response = asr_client.recognize_file(dummy_audio_file, config)

    # Print results
    if response.results:
        for result in response.results:
            if result.alternatives:
                print(f"Transcription: {result.alternatives[0].transcript}")
            else:
                print("No alternatives found for this segment.")
    else:
        print("No speech recognized in the audio.")

except Exception as e:
    print(f"An error occurred: {e}")
    print("Please ensure the Riva server is running and accessible at the specified RIVA_URI.")

view raw JSON →