Riva SR Plugin 1.0.0 Released

[Arsen Chaloyan]

Riva Speech Recognition (SR) Plugin 1.0.0 to the UniMRCP Server (UMS) has been released.

The plugin is based on the following components:

• UniMRCP Server 1.7.0
• NVIDIA Riva Speech Skills 1.9.0-beta
• gRPC 1.30.3
• Protobuf 3.12.2

The binaries are currently available for the following Linux distributions:

• Red Hat / CentOS 7 (unimrcp-riva-sr-1.7.0-1.el7.x86_64.rpm)
• Red Hat / CentOS 8 (unimrcp-riva-sr-1.7.0-1.el8.x86_64.rpm)
• Ubuntu 18.04 LTS (unimrcp-riva-sr_1.7.0-bionic_amd64.deb)
• Ubuntu 20.04 LTS (unimrcp-riva-sr_1.7.0-focal_amd64.deb)

IVR platforms can now utilize NVIDIA Riva Speech-to-Text API via UniMRCP Server.

NVIDIA Riva Speech-to-Text API performs speech to text conversion powered by machine learning providing the following main features.

Streaming Speech Recognition

Supports efficient streaming speech transcription.

Low Latency

Intermediate transcripts are returned with low latency.

Efficient Feature Extraction

GPU-accelerated feature extraction.

Multiple Acoustic Models

Multiple (and growing) acoustic model architecture options accelerated by NVIDIA TensorRT

Beam Search Decoder

Beam search decoder based on n-gram language models

Voice Activity Detection

CTC-based voice activity detection algorithms.

Automatic Punctuation

Automatic punctuation can optionally be enabled.

Alternate Transcripts

Ability to return top-N transcripts from beam decoder

Word-level Timestamps

Word-level timestamps can optionally be returned.

Inverse Text Normalization

Inverse text normalization (ITN) is supported.

For installation, configuration and usage instructions, follow the guides available on a new Riva SR plugin page.

• https://unispeech.io/rivasr
  
• https://www.unimrcp.org/rivasr
  

Please note this is a commercial module. Everyone is welcome to start evaluation by requesting a trial license.

Thank you for using UniMRCP.

--

Arsen Chaloyan
Author of UniMRCP
http://www.unimrcp.org