wav2vec2-xls-r-1b-ka Open-source Automatic Speech Recognition Model - Free Deployment for Georgian Speech Recognition

Home

Wav2vec2 Xls R 1b Ka

Developed by arampacha

An automatic speech recognition model fine-tuned on Georgian language datasets based on facebook/wav2vec2-xls-r-1b

Speech Recognition

Transformers

OtherOpen Source License:Apache-2.0 #Georgian speech recognition #Low word error rate #Multi-scenario adaptation

Downloads 25

Release Time : 3/2/2022

Model Overview

This is an automatic speech recognition (ASR) model optimized for Georgian language, demonstrating excellent performance on Mozilla Common Voice and Robust Speech Event datasets

Model Features

High-performance speech recognition

Achieves low word error rate (WER) and character error rate (CER) in Georgian speech recognition tasks

Large-scale pretraining foundation

Fine-tuned from the 1-billion parameter wav2vec2-xls-r model with powerful speech feature extraction capabilities

Multi-dataset validation

Performance validated on multiple datasets including Common Voice and Robust Speech Events

Model Capabilities

Georgian speech recognition

Speech-to-text

Robust speech processing

Use Cases

Speech transcription

Georgian speech transcription

Convert Georgian speech content into text

Word error rate 7.4%, character error rate 1.19%

Voice assistant

Georgian voice interaction

Supports Georgian voice command recognition

Word error rate 21.58-22.61% on Robust Speech Event dataset

🚀 wav2vec2-xls-r-1b-ka

This model is a fine - tuned speech recognition model. It is based on facebook/wav2vec2-xls-r-1b and trained on the /WORKSPACE/DATA/KA/NOIZY_STUDENT_2/ - KA dataset. It has achieved good results on the evaluation set, which can be used for automatic speech recognition tasks.

📄 License

This model is under the Apache - 2.0 license.

📋 Model Information

Property	Details
Language	ka
Tags	automatic - speech - recognition, mozilla - foundation/common_voice_8_0, generated_from_trainer, robust - speech - event, hf - asr - leaderboard
Datasets	common_voice

📊 Model Results

The model has the following performance on different datasets:

Common Voice ka

Task: Speech Recognition
Metrics:
- WER LM: 7.39778066580026
- CER LM: 1.1882089427096434

Robust Speech Event - Dev Data

Task: Automatic Speech Recognition
Metrics:
- Test WER: 22.61

Robust Speech Event - Test Data

Task: Automatic Speech Recognition
Metrics:
- Test WER: 21.58

📈 Evaluation Results

On the evaluation set, the model achieves the following results:

Loss: 0.1022
Wer: 0.1527
Cer: 0.0221

🔧 Technical Details

Training Hyperparameters

The following hyperparameters were used during training:

learning_rate: 7e - 05
train_batch_size: 16
eval_batch_size: 64
seed: 42
gradient_accumulation_steps: 8
total_train_batch_size: 128
optimizer: Adam with betas=(0.9,0.98) and epsilon = 1e - 08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
training_steps: 4000
mixed_precision_training: Native AMP

Training Results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.2839	6.45	400	0.2229	0.3609	0.0557
0.9775	12.9	800	0.1271	0.2202	0.0317
0.9045	19.35	1200	0.1268	0.2030	0.0294
0.8652	25.8	1600	0.1211	0.1940	0.0287
0.8505	32.26	2000	0.1192	0.1912	0.0276
0.8168	38.7	2400	0.1086	0.1763	0.0260
0.7737	45.16	2800	0.1098	0.1753	0.0256
0.744	51.61	3200	0.1054	0.1646	0.0239
0.7114	58.06	3600	0.1034	0.1573	0.0228
0.6773	64.51	4000	0.1022	0.1527	0.0221

Framework Versions

Transformers 4.17.0.dev0
Pytorch 1.10.2
Datasets 1.18.4.dev0
Tokenizers 0.11.0

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご