Open-source Automatic Speech Recognition Model - Supports Precise Portuguese Speech Recognition Based on Fine-tuning

Output

Developed by tonyalves

Automatic speech recognition model fine-tuned on Mozilla Common Voice Portuguese dataset based on facebook/wav2vec2-xls-r-300m

Speech Recognition

Transformers

OtherOpen Source License:Apache-2.0 #Portuguese speech recognition #Wav2Vec2 fine-tuning #Low word error rate

Downloads 28

Release Time : 4/1/2022

Model Overview

This model is an automatic speech recognition (ASR) model for Portuguese, capable of converting Portuguese speech into text.

Model Features

High accuracy

Achieved a word error rate (WER) of 13.52% on the evaluation set

Based on large-scale pre-trained model

Fine-tuned from facebook/wav2vec2-xls-r-300m model, inheriting its powerful speech feature extraction capabilities

Portuguese optimization

Specially optimized for Portuguese speech data, suitable for Portuguese speech recognition tasks

Model Capabilities

Portuguese speech recognition

Speech-to-text

Use Cases

Speech transcription

Voice memo transcription

Convert Portuguese voice memos into searchable text

Assistive technology

Voice control applications

Provide voice control interfaces for Portuguese users

🚀 Output

This model is a fine - tuned version of [facebook/wav2vec2 - xls - r - 300m](https://huggingface.co/facebook/wav2vec2 - xls - r - 300m) on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - PT dataset. It offers a solution for automatic speech recognition tasks, leveraging the pre - trained model and fine - tuning it on specific data to improve performance.

It achieves the following results on the evaluation set:

Loss: 0.1505
Wer: 0.1352

📚 Documentation

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

🔧 Technical Details

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 7.5e - 05
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 2000
num_epochs: 50.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.1367	0.64	500	3.8825	1.0
2.9677	1.29	1000	2.9498	1.0
1.5884	1.93	1500	0.6722	0.6493
1.2292	2.57	2000	0.3635	0.3202
1.1314	3.22	2500	0.2970	0.2680
1.0879	3.86	3000	0.2671	0.2486
1.0344	4.5	3500	0.2625	0.2239
1.0109	5.15	4000	0.2520	0.2230
0.9966	5.79	4500	0.2280	0.2105
0.9815	6.43	5000	0.2254	0.2179
0.9744	7.08	5500	0.2301	0.2137
0.9487	7.72	6000	0.2224	0.2051
0.9431	8.37	6500	0.2105	0.1992
0.9365	9.01	7000	0.2114	0.2019
0.9268	9.65	7500	0.2097	0.1988
0.9292	10.3	8000	0.2120	0.1986
0.929	10.94	8500	0.2048	0.1998
0.9017	11.58	9000	0.2035	0.1999
0.8898	12.23	9500	0.1961	0.1908
0.8799	12.87	10000	0.1945	0.1817
0.869	13.51	10500	0.1929	0.1844
0.8572	14.16	11000	0.1941	0.1888
0.8691	14.8	11500	0.1912	0.1804
0.8645	15.44	12000	0.1950	0.1851
0.8468	16.09	12500	0.1879	0.1770
0.8405	16.73	13000	0.1881	0.1759
0.8647	17.37	13500	0.1861	0.1740
0.8477	18.02	14000	0.1782	0.1702
0.811	18.66	14500	0.1915	0.1757
0.8165	19.3	15000	0.1820	0.1724
0.8166	19.95	15500	0.1798	0.1697
0.8167	20.59	16000	0.1805	0.1752
0.7908	21.24	16500	0.1761	0.1699
0.7925	21.88	17000	0.1740	0.1709
0.7803	22.52	17500	0.1815	0.1727
0.7839	23.17	18000	0.1737	0.1694
0.7815	23.81	18500	0.1732	0.1630
0.767	24.45	19000	0.1724	0.1648
0.7672	25.1	19500	0.1706	0.1596
0.7691	25.74	20000	0.1718	0.1618
0.7547	26.38	20500	0.1694	0.1565
0.7498	27.03	21000	0.1706	0.1582
0.7459	27.67	21500	0.1663	0.1586
0.7374	28.31	22000	0.1651	0.1567
0.7499	28.96	22500	0.1668	0.1549
0.7471	29.6	23000	0.1667	0.1553
0.7369	30.24	23500	0.1659	0.1556
0.7389	30.89	24000	0.1668	0.1538
0.7197	31.53	24500	0.1687	0.1561
0.71	32.17	25000	0.1666	0.1516
0.7199	32.82	25500	0.1640	0.1523
0.7194	33.46	26000	0.1659	0.1528
0.6923	34.11	26500	0.1662	0.1507
0.7054	34.75	27000	0.1641	0.1486
0.6955	35.39	27500	0.1634	0.1497
0.7084	36.04	28000	0.1618	0.1478
0.6917	36.68	28500	0.1589	0.1471
0.687	37.32	29000	0.1589	0.1450
0.6914	37.97	29500	0.1588	0.1465
0.6646	38.61	30000	0.1602	0.1468
0.6667	39.25	30500	0.1588	0.1444
0.6754	39.9	31000	0.1587	0.1455
0.6632	40.54	31500	0.1586	0.1461
0.6619	41.18	32000	0.1571	0.1441
0.6561	41.83	32500	0.1564	0.1420
0.6492	42.47	33000	0.1539	0.1437
0.6649	43.11	33500	0.1512	0.1406
0.6511	43.76	34000	0.1539	0.1384
0.6551	44.4	34500	0.1520	0.1384
0.6452	45.05	35000	0.1510	0.1368
0.6155	45.69	35500	0.1522	0.1375
0.628	46.33	36000	0.1522	0.1366
0.6389	46.97	36500	0.1513	0.1377
0.6265	47.62	37000	0.1512	0.1369
0.6197	48.26	37500	0.1511	0.1362
0.621	48.91	38000	0.1510	0.1357
0.6259	49.55	38500	0.1506	0.1353

Framework versions

Transformers 4.17.0
Pytorch 1.9.1+cu102
Datasets 2.0.0
Tokenizers 0.11.6

📄 License

This project 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