Open-source Automatic Speech Recognition Model XLS-R-EUS - Free Deployment, Precise Recognition of Basque Language Speech

Xls R Eus

Developed by shpotes

This is an automatic speech recognition model fine-tuned on the Common Voice 8 Basque dataset based on facebook/wav2vec2-xls-r-300m

Speech Recognition

Transformers

OtherOpen Source License:Apache-2.0 #Basque speech recognition #Low word error rate #Multilingual support

Downloads 14

Release Time : 3/2/2022

Model Overview

This model is specifically designed for Basque speech-to-text tasks, achieving a word error rate (WER) of 17.87% and a character error rate (CER) of 3.26% on the Common Voice 8 Basque test set

Model Features

High-performance Basque recognition

Achieves 17.87% word error rate on the Common Voice 8 Basque test set

Based on XLS-R architecture

Uses facebook's wav2vec2-xls-r-300m as the base model

Optimized training process

Employs advanced training techniques like cosine annealing learning rate scheduling and mixed-precision training

Model Capabilities

Basque speech recognition

Speech-to-text

Robust speech processing

Use Cases

Speech transcription

Basque speech transcription

Convert Basque speech content into text

17.87% word error rate, 3.26% character error rate

Voice assistants

Basque voice assistant

Provide voice interaction capabilities for Basque users

🚀 XLS-R-EUS Model

This model, xls-r-eus, is a fine - tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - EU dataset. It offers high - performance in automatic speech recognition tasks.

🚀 Quick Start

This model is a fine - tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - EU dataset. It achieves the following results on the evaluation set:

Loss: 0.2278
Wer: 0.1787

✨ Features

Fine - tuned: Based on the pre - trained facebook/wav2vec2-xls-r-300m, it is fine - tuned on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - EU dataset.
High Performance: Achieves low WER (Word Error Rate) and CER (Character Error Rate) on the evaluation set.

📦 Installation

No specific installation steps are provided in the original document.

📚 Documentation

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0003
train_batch_size: 72
eval_batch_size: 72
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 144
optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 500
num_epochs: 100.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2548	4.24	500	0.2470	0.3663
0.1435	8.47	1000	0.2000	0.2791
0.1158	12.71	1500	0.2030	0.2652
0.1094	16.95	2000	0.2096	0.2605
0.1004	21.19	2500	0.2150	0.2477
0.0945	25.42	3000	0.2072	0.2369
0.0844	29.66	3500	0.1981	0.2328
0.0877	33.89	4000	0.2041	0.2425
0.0741	38.14	4500	0.2353	0.2421
0.0676	42.37	5000	0.2092	0.2213
0.0623	46.61	5500	0.2217	0.2250
0.0574	50.84	6000	0.2152	0.2179
0.0583	55.08	6500	0.2207	0.2186
0.0488	59.32	7000	0.2225	0.2159
0.0456	63.56	7500	0.2293	0.2031
0.041	67.79	8000	0.2277	0.2013
0.0379	72.03	8500	0.2287	0.1991
0.0381	76.27	9000	0.2233	0.1954
0.0308	80.51	9500	0.2195	0.1835
0.0291	84.74	10000	0.2266	0.1825
0.0266	88.98	10500	0.2285	0.1801
0.0266	93.22	11000	0.2292	0.1801
0.0262	97.46	11500	0.2278	0.1788

Framework versions

Transformers 4.16.0.dev0
Pytorch 1.10.1+cu102
Datasets 1.18.4.dev0
Tokenizers 0.11.0

📄 License

This model is released under the Apache - 2.0 license.

📊 Model Information

Property	Details
Model Type	Fine - tuned version of facebook/wav2vec2-xls-r-300m
Training Data	MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - EU dataset
Test WER	0.17871523648578164
Test CER	0.032624506085144

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