wav2vec2-large-xls-r-300m-guarani-small-wb open-source model - Free implementation of automatic Guaraní speech recognition

Wav2vec2 Large Xls R 300m Guarani Small Wb

Developed by jhonparra18

This model is an automatic speech recognition (ASR) model based on the wav2vec2-large-xls-r-300m architecture, fine-tuned on the Guarani speech dataset.

Speech Recognition

Transformers

Open Source License:Apache-2.0 #Guarani speech recognition #Low-resource language model #XLS-R architecture

Downloads 16

Release Time : 4/25/2022

Model Overview

This is a speech recognition model optimized for Guarani, based on Facebook's wav2vec2-large-xls-r-300m architecture and fine-tuned on the Common Voice dataset.

Model Features

Low-resource language support

Specifically optimized for low-resource languages like Guarani

Efficient fine-tuning

Targeted fine-tuning based on pre-trained large-scale multilingual models

Good performance

Achieved a word error rate of 24.46% and a character error rate of 3.68% on the evaluation set

Model Capabilities

Speech-to-text

Guarani speech recognition

Use Cases

Speech transcription

Guarani speech transcription

Convert Guarani speech content into text

Word error rate 24.46%, character error rate 3.68%

Language preservation

Minority language digitization

Help preserve and digitize minority languages like Guarani

🚀 wav2vec2-large-xls-r-300m-guarani-small-wb

This model is a fine - tuned version of [glob - asr/wav2vec2 - large - xls - r - 300m - guarani - small](https://huggingface.co/glob - asr/wav2vec2 - large - xls - r - 300m - guarani - small) on the common_voice dataset, aiming to achieve better performance in speech - related tasks.

🚀 Quick Start

This model is a fine - tuned version of [glob - asr/wav2vec2 - large - xls - r - 300m - guarani - small](https://huggingface.co/glob - asr/wav2vec2 - large - xls - r - 300m - guarani - small) on the common_voice dataset. It achieves the following results on the evaluation set:

Loss: 0.1622
Wer: 0.2446
Cer: 0.0368

📚 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: 0.0002
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 16
optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 10
num_epochs: 20
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.1818	0.32	10	0.1196	0.2146	0.0305
0.2953	0.65	20	0.1801	0.3090	0.0426
0.2941	0.97	30	0.1935	0.3090	0.0420
0.2786	1.29	40	0.1899	0.3305	0.0483
0.2665	1.61	50	0.1716	0.3176	0.0454
0.2752	1.94	60	0.1895	0.3948	0.0564
0.2482	2.26	70	0.1753	0.3176	0.0449
0.2486	2.58	80	0.1501	0.2747	0.0403
0.2878	2.9	90	0.1890	0.3348	0.0529
0.2539	3.23	100	0.2076	0.4635	0.0610
0.2069	3.55	110	0.1711	0.3476	0.0466
0.2262	3.87	120	0.1839	0.3605	0.0500
0.2032	4.19	130	0.1724	0.3391	0.0489
0.1997	4.52	140	0.1498	0.2704	0.0414
0.2216	4.84	150	0.1531	0.3047	0.0472
0.2294	5.16	160	0.1882	0.3176	0.0500
0.2305	5.48	170	0.1799	0.3176	0.0483
0.2052	5.81	180	0.1645	0.3262	0.0477
0.2192	6.13	190	0.1439	0.2060	0.0339
0.1844	6.45	200	0.1557	0.2918	0.0403
0.1803	6.77	210	0.1664	0.3004	0.0426
0.1831	7.1	220	0.1780	0.3176	0.0477
0.1618	7.42	230	0.1671	0.2661	0.0437
0.1528	7.74	240	0.2108	0.3176	0.0506
0.1335	8.06	250	0.1677	0.2575	0.0408
0.1736	8.39	260	0.1581	0.3004	0.0460
0.1607	8.71	270	0.1529	0.3047	0.0403
0.1451	9.03	280	0.1666	0.2747	0.0408
0.1534	9.35	290	0.1722	0.2833	0.0437
0.1567	9.68	300	0.1747	0.2918	0.0397
0.1356	10.0	310	0.1659	0.2961	0.0443
0.1248	10.32	320	0.1752	0.3348	0.0449
0.149	10.65	330	0.1792	0.3348	0.0449
0.1471	10.97	340	0.1843	0.3391	0.0460
0.1564	11.29	350	0.2015	0.3433	0.0460
0.1597	11.61	360	0.1798	0.2618	0.0380
0.161	11.94	370	0.1716	0.2747	0.0374
0.1481	12.26	380	0.1776	0.2747	0.0397
0.1168	12.58	390	0.1900	0.2961	0.0454
0.1173	12.9	400	0.1987	0.3090	0.0454
0.1245	13.23	410	0.1710	0.2918	0.0408
0.1118	13.55	420	0.1808	0.3047	0.0431
0.1111	13.87	430	0.1893	0.2747	0.0403
0.1041	14.19	440	0.1876	0.2918	0.0431
0.1152	14.52	450	0.1800	0.2790	0.0408
0.107	14.84	460	0.1717	0.2747	0.0385
0.1139	15.16	470	0.1652	0.2704	0.0391
0.0922	15.48	480	0.1659	0.2618	0.0391
0.101	15.81	490	0.1610	0.2489	0.0362
0.0835	16.13	500	0.1584	0.2403	0.0362
0.1251	16.45	510	0.1601	0.2575	0.0380
0.0888	16.77	520	0.1632	0.2661	0.0380
0.0968	17.1	530	0.1674	0.2661	0.0385
0.1105	17.42	540	0.1629	0.2833	0.0391
0.0914	17.74	550	0.1623	0.3090	0.0408
0.0843	18.06	560	0.1611	0.3004	0.0408
0.0861	18.39	570	0.1583	0.2661	0.0385
0.0861	18.71	580	0.1579	0.2618	0.0385
0.0678	19.03	590	0.1585	0.2661	0.0374
0.0934	19.35	600	0.1613	0.2489	0.0368
0.0976	19.68	610	0.1617	0.2446	0.0368
0.0799	20.0	620	0.1622	0.2446	0.0368

Framework versions

Transformers 4.18.0
Pytorch 1.11.0+cu113
Datasets 2.1.0
Tokenizers 0.12.1

📄 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