wav2vec2-xls-r-300m-hy Open Source Model - Free and Accurate Armenian Speech-to-Text Conversion

Wav2vec2 Xls R 300m Hy

Developed by arampacha

An automatic speech recognition (ASR) model fine-tuned on Armenian language datasets based on facebook/wav2vec2-xls-r-300m, supporting Armenian speech-to-text tasks.

Speech Recognition

Transformers

OtherOpen Source License:Apache-2.0 #Armenian speech recognition #Low character error rate #Multi-scenario robustness

Downloads 25

Release Time : 3/2/2022

Model Overview

This model is specifically optimized for Armenian speech recognition, achieving a word error rate (WER) of 13.19% and a character error rate (CER) of 2.79% on the Common Voice hy-AM dataset.

Model Features

Low Word Error Rate

Achieves 13.19% word error rate (WER) on the Common Voice hy-AM test set

Low Character Error Rate

Achieves 2.79% character error rate (CER) on the Common Voice hy-AM test set

Armenian Language Optimization

Specially fine-tuned and optimized for Armenian speech characteristics

Based on XLS-R Architecture

Utilizes Facebook's advanced wav2vec2-xls-r-300m architecture

Model Capabilities

Armenian speech recognition

Speech-to-text

Robust speech processing

Use Cases

Speech Transcription

Armenian Speech Transcription

Convert Armenian speech content into text

Word error rate 13.19%, character error rate 2.79%

Voice Assistants

Armenian Voice Command Recognition

Used for voice command recognition in Armenian voice assistants or control systems

🚀 wav2vec2-xls-r-300m-hy

This is a fine - tuned model for Armenian automatic speech recognition, achieving good results on relevant datasets.

🚀 Quick Start

This model is a fine - tuned version of facebook/wav2vec2-xls-r-300m on the /WORKSPACE/DATA/HY/NOIZY_STUDENT_3/ - NA dataset. It achieves the following results on the evaluation set:

Loss: 0.2293
Wer: 0.3333
Cer: 0.0602

✨ Features

Model Information

Property	Details
Model Type	wav2vec2-xls-r-300m-hy
Tags	automatic - speech - recognition, mozilla - foundation/common_voice_8_0, generated_from_trainer, robust - speech - event, hy, hf - asr - leaderboard
Datasets	common_voice
License	apache - 2.0

Evaluation Results

Task	Dataset	Metrics	Value
Automatic Speech Recognition	Common Voice hy - AM (mozilla - foundation/common_voice_8_0)	WER LM	13.192818110850899
Automatic Speech Recognition	Common Voice hy - AM (mozilla - foundation/common_voice_8_0)	CER LM	2.787051087506323
Automatic Speech Recognition	Robust Speech Event - Dev Data (speech - recognition - community - v2/dev_data)	Test WER	22.246048764990867
Automatic Speech Recognition	Robust Speech Event - Dev Data (speech - recognition - community - v2/dev_data)	Test CER	7.59406739840239

🔧 Technical Details

Training Hyperparameters

The following hyperparameters were used during training:

learning_rate: 7e - 05
train_batch_size: 64
eval_batch_size: 64
seed: 842
gradient_accumulation_steps: 2
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
3.1471	7.02	400	3.1599	1.0	1.0
1.8691	14.04	800	0.7674	0.7361	0.1686
1.3227	21.05	1200	0.3849	0.5336	0.1007
1.163	28.07	1600	0.3015	0.4559	0.0823
1.0768	35.09	2000	0.2721	0.4032	0.0728
1.0224	42.11	2400	0.2586	0.3825	0.0691
0.9817	49.12	2800	0.2458	0.3653	0.0653
0.941	56.14	3200	0.2306	0.3388	0.0605
0.9235	63.16	3600	0.2315	0.3380	0.0615
0.9141	70.18	4000	0.2293	0.3333	0.0602

Framework Versions

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

📄 License

This model is licensed under the apache - 2.0 license.

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