W

Wav2vec2 Large Ru Golos With Lm

Developed by bond005
This is a Russian speech recognition model fine-tuned from facebook/wav2vec2-large-xlsr-53, trained on the Sberdevices Golos dataset and integrated with a 2-gram language model to improve recognition accuracy.
Downloads 434
Release Time : 9/26/2022

Model Overview

This model is specifically designed for Russian speech recognition tasks, supporting 16kHz audio input and demonstrating excellent performance on multiple Russian test sets.

Model Features

Integrated Language Model
Incorporates a 2-gram language model built on Russian text corpora, significantly improving recognition accuracy.
Data Augmentation Training
Applied audio enhancement techniques such as pitch shifting, speed variation, and reverberation during training to enhance model robustness.
Multi-dataset Evaluation
Comprehensively evaluated on multiple test sets including Sberdevices Golos and Common Voice Russian.

Model Capabilities

Russian speech recognition
Audio transcription
Speech-to-text

Use Cases

Voice Assistants
Smart Home Control
Used for recognizing voice commands for Russian smart home devices.
Achieved a CER of 5.128% on far-field test sets.
Speech Transcription
Meeting Minutes Transcription
Automatically transcribes Russian meeting recordings into text.
Achieved a WER of 6.883% on crowdsourced test sets.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase