Multilingual ASR across 25 European languages with punctuation, capitalization, and word timestamps
Riva supports gRPC, HTTP and Realtime APIs.
Inference using HTTP API. Replace <path_to_audio_file> with the full path of the wav file to be transcribed. This model serves 25 European locales — pass the BCP-47 code matching your audio (e.g. en-GB, de-DE, fr-FR, es-ES).
curl -s https://None.invocation.api.nvcf.nvidia.com/v1/audio/transcriptions \
-H "Authorization: Bearer $NVIDIA_API_KEY" \
-F language=en-GB \
-F file="@<path_to_audio_file>"
Instructions below demonstrate usage of parakeet-tdt-0_6b model using Python gRPC client. For full API details refer gRPC API Reference.
You will need a system with Git and Python 3+ installed.
pip install -U nvidia-riva-client
Download Python client code by cloning Python Client Repository.
git clone https://github.com/nvidia-riva/python-clients.git
Open a command terminal and execute below command to transcribe audio. Make sure you have a speech file in 16-bit Mono format in WAV/OGG/OPUS container. If you have generated the API key, it will be auto-populated in the command.
The model auto-detects the spoken language across 25 supported European languages. Pass the appropriate BCP-47 language code via --language-code (e.g. en-GB, de-DE, fr-FR, es-ES).
python python-clients/scripts/asr/transcribe_file_offline.py \
--server grpc.nvcf.nvidia.com:443 --use-ssl \
--metadata function-id "None" \
--metadata "authorization" "Bearer $NVIDIA_API_KEY" \
--language-code en-GB \
--word-time-offsets --automatic-punctuation \
--input-file <path_to_audio_file>
Riva uses gRPC APIs. Proto files can be downloaded from Riva gRPC Proto files and compiled to target language using Protoc compiler. Example Riva clients in C++ and Python languages are provided below.