S

Stt Uk Citrinet 1024 Gamma 0 25

Developed by nvidia
This is a streaming Citrinet model for Ukrainian automatic speech recognition (ASR) with 141 million parameters, trained on 69 hours of Ukrainian speech data, achieving a test WER as low as 3.52%.
Downloads 65
Release Time : 7/27/2022

Model Overview

This model is a non-autoregressive variant of streaming Citrinet using CTC loss/decoding, capable of transcribing Ukrainian lowercase speech including spaces and apostrophes.

Model Features

Cross-lingual transfer learning
This model was fine-tuned from a pre-trained Russian Citrinet-1024 model through cross-lingual transfer learning
High performance
Achieves excellent WER performance across multiple versions of Mozilla Common Voice test sets, with the lowest reaching 3.52%
Streaming processing
Supports streaming speech recognition, suitable for real-time applications
Riva compatible
Compatible with NVIDIA Riva for production-level server deployment

Model Capabilities

Ukrainian speech recognition
Real-time speech transcription
Batch processing of audio files

Use Cases

Speech transcription
Speech-to-text service
Convert Ukrainian speech content into text
High-accuracy transcription with WER as low as 3.52%
Real-time applications
Real-time caption generation
Generate real-time captions for Ukrainian videos or live streams
Streaming capability supports low-latency applications
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase