NVIDIA Riva Clients

NVIDIA Riva is a GPU-accelerated SDK for building Speech AI applications that are customized for your use case and deliver real-time performance. This repo provides performant client example command-line clients.

Main API

• riva.client.ASRService is a class for speech recognition,
• riva.client.TTSService is a class for speech synthesis,
• riva.client.NLPService is a class for natural language processing.

CLI interface

• Automatic Speech Recognition (ASR)
  • scripts/asr/riva_streaming_asr_client.py demonstrates streaming transcription in several threads, can prints time stamps.
  • scripts/asr/transcribe_file.py performs streaming transcription,
  • scripts/asr/transcribe_file_offline.py performs offline transcription,
  • scripts/asr/transcribe_mic.py performs streaming transcription of audio acquired through microphone.
• Speech Synthesis (TTS)
  • scripts/tts/talk.py synthesizes audio for a text in streaming or offline mode.
• Natural Language Processing (NLP)
  • scripts/nlp/intentslot_client.py recognizes intents and slots in input sentences,
  • scripts/nlp/ner_client.py detects named entities in input sentences,
  • scripts/nlp/punctuation_client.py restores punctuation and capitalization in input sentences,
  • scripts/nlp/qa_client.py queries a document with natural language query and prints answer from a document,
  • scripts/nlp/text_classify_client.py classifies input sentences,
  • scripts/nlp/eval_intent_slot.py prints intents and slots classification reports for test data.

Installation

1. Create a conda environment and activate it
2. From source:
   • Clone riva-python-clients repo and change to the repo root
   • Run commands

git clone https://github.com/nvidia-riva/python-clients.git
cd python-clients
git submodule init
git submodule update --remote --recursive
pip install -r requirements.txt
python3 setup.py bdist_wheel
pip install --force-reinstall dist/*.whl

3. pip:

pip install nvidia-riva-client

If you would like to use output and input audio devices (scripts scripts/asr/transcribe_file_rt.py, scripts/asr/transcribe_mic.py, scripts/tts/talk.py or module riva.client/audio_io.py), you will need to install PyAudio.

conda install -c anaconda pyaudio

For NLP evaluation you will need transformers and sklearn libraries.

pip install -U scikit-learn
pip install -U transformers

Before using microphone and audio output devices on Unix

you may need to run commands

adduser $USER audio
adduser $USER pulse-access

and restart.

Usage

Server

Before running client part of Riva, please set up a server. The simplest way to do this is to follow quick start guide.

CLI

You may find all CLI scripts in scripts directory. Each script has a description of its purpose and parameters.

ASR

You may find a detailed documentation here.

For transcribing in streaming mode you may use scripts/asr/transcribe_file.py.

python scripts/asr/transcribe_file.py \
    --input-file data/examples/en-US_AntiBERTa_for_word_boosting_testing.wav

You may watch how a transcript grows if you set --simulate-realtime and --show-intermediate.

python scripts/asr/transcribe_file.py \
    --input-file data/examples/en-US_AntiBERTa_for_word_boosting_testing.wav \
    --simulate-realtime \
    --show-intermediate

You may listen audio simultaneously with transcribing (you will need installed PyAudio and access to audio devices).

python scripts/asr/transcribe_file.py \
    --input-file data/examples/en-US_AntiBERTa_for_word_boosting_testing.wav \
    --play-audio \
    --show-intermediate

Offline transcription is performed this way.

python scripts/asr/transcribe_file_offline.py \
    --input-file data/examples/en-US_AntiBERTa_for_word_boosting_testing.wav

You can improve transcription of this audio by word boosting.

python scripts/asr/transcribe_file_offline.py \
  --input-file data/examples/en-US_AntiBERTa_for_word_boosting_testing.wav \
  --boosted-lm-words AntiBERTa \
  --boosted-lm-words ABlooper \
  --boosted-lm-score 20.0

NLP

You can provide inputs to scripts/nlp/intentslot_client.py, scripts/nlp/punctuation_client.py both through command line arguments and interactively.

python scripts/nlp/intentslot_client.py --query "What is the weather tomorrow?"

or

python scripts/nlp/intentslot_client.py --interactive

For punctuation client the commands look similar.

python scripts/nlp/punctuation_client.py --query "can you prove that you are self aware"

or

python scripts/nlp/punctuation_client.py --interactive

NER client can output 1 of the following: label name, span start, span end

python scripts/nlp/ner_client.py \
  --query "Where is San Francisco?" "Jensen Huang is the CEO of NVIDIA Corporation." \
  --test label

or

python scripts/nlp/ner_client.py \
  --query "Where is San Francisco?" "Jensen Huang is the CEO of NVIDIA Corporation." \
  --test span_start

or

python scripts/nlp/ner_client.py \
  --query "Where is San Francisco?" "Jensen Huang is the CEO of NVIDIA Corporation." \
  --test span_end

Provide query and context to QA client.

python scripts/nlp/qa_client.py \
  --query "How many gigatons of carbon dioxide was released in 2005?" \
  --context "In 2010 the Amazon rainforest experienced another severe drought, in some ways "\
"more extreme than the 2005 drought. The affected region was approximate 1,160,000 square "\
"miles (3,000,000 km2) of rainforest, compared to 734,000 square miles (1,900,000 km2) in "\
"2005. The 2010 drought had three epicenters where vegetation died off, whereas in 2005 the "\
"drought was focused on the southwestern part. The findings were published in the journal "\
"Science. In a typical year the Amazon absorbs 1.5 gigatons of carbon dioxide; during 2005 "\
"instead 5 gigatons were released and in 2010 8 gigatons were released."

Text classification requires only a query.

python scripts/nlp/text_classify_client.py --query "How much sun does california get?"

TTS

Call scripts/tts/talk.py script, and you will be prompted to enter a text for speech synthesis. Set --play-audio option, and a synthesized speech will be played.

python scripts/tts/talk.py --play-audio

You can write output to file.

python scripts/tts/talk.py --output 'my_synth_speech.wav'

You can use streaming mode (audio fragments returned to client as soon as they are ready).

python scripts/tts/talk.py --stream --play-audio

API

See tutorial notebooks in directory tutorials.

Documentation

Additional documentation on the Riva Speech Skills SDK can be found here.

License

This client code is MIT-licensed. See LICENSE file for full details.