S

Stt Ru Fastconformer Hybrid Large Pc

Developed by nvidia
This is a FastConformer hybrid model for Russian automatic speech recognition, combining Transducer and CTC decoders with approximately 115 million parameters.
Downloads 6,513
Release Time : 5/26/2023

Model Overview

The model can transcribe speech containing uppercase and lowercase Russian letters, spaces, and basic punctuation marks, suitable for Russian speech recognition tasks.

Model Features

Hybrid Training Architecture
Utilizes both Transducer and CTC loss functions during training to enhance model robustness
Optimized FastConformer
Employs an optimized Conformer architecture with 8x depthwise separable convolution downsampling for improved processing efficiency
Multi-dataset Training
Trained on a composite dataset containing 1,840 hours of Russian speech, covering various speech scenarios

Model Capabilities

Russian speech recognition
Punctuation prediction
Case sensitivity recognition

Use Cases

Speech Transcription
Russian Speech-to-Text
Convert Russian speech content into text format
Achieves a WER of 5.3 on the Common Voice 12.0 test set
Voice Assistants
Russian Voice Command Recognition
Recognize and understand Russian voice commands
Achieves a WER as low as 1.9 on the Golos crowd test set
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase