victor-hg-ptbr-2.0 Open-Source Portuguese Speech Recognition Model

Home

Victor Hg Ptbr 2.0

Developed by Vkt

Portuguese speech recognition model fine-tuned on the common_voice dataset based on facebook/wav2vec2-xls-r-300m

Speech Recognition

Transformers

Open Source License:Apache-2.0 #Portuguese speech recognition #Low word error rate #XLS-R architecture

Downloads 30

Release Time : 5/24/2022

Model Overview

This is an automatic speech recognition (ASR) model optimized for Portuguese, fine-tuned on the XLS-R architecture, suitable for Brazilian Portuguese speech-to-text tasks.

Model Features

High accuracy

Achieves a word error rate (WER) of 2.19% on the evaluation set, demonstrating excellent performance

Large-scale pre-training foundation

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

Optimized for Brazilian Portuguese

Specifically fine-tuned for Brazilian Portuguese, adapting to regional linguistic characteristics

Model Capabilities

Brazilian Portuguese speech recognition

Audio-to-text conversion

Speech transcription

Use Cases

Speech transcription

Automated meeting minutes

Automatically convert Portuguese meeting recordings into text transcripts

Highly accurate transcriptions

Voice assistants

Build Portuguese voice assistants for the Brazilian market

Accurate voice command recognition

Accessibility technology

Real-time caption generation

Generate real-time captions for Portuguese video content

Improved content accessibility

🚀 victor-hg-ptbr-2.0

This model is a fine - tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It can achieve high - quality speech recognition results with low loss and word error rate.

🚀 Quick Start

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

Loss: 0.0240
Wer: 0.0219

📚 Documentation

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: 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: 500
num_epochs: 30
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.4069	0.21	400	1.1372	0.9140
0.8079	0.43	800	0.5822	0.5339
0.5821	0.64	1200	0.4226	0.4177
0.5159	0.86	1600	0.4074	0.3970
0.4484	1.07	2000	0.3144	0.3220
0.3937	1.29	2400	0.3160	0.3264
0.3911	1.5	2800	0.2863	0.2956
0.3761	1.71	3200	0.3029	0.3128
0.3722	1.93	3600	0.2771	0.2933
0.3193	2.14	4000	0.2603	0.2795
0.3013	2.36	4400	0.2682	0.2703
0.3039	2.57	4800	0.2630	0.2618
0.3133	2.79	5200	0.2578	0.2629
0.3173	3.0	5600	0.2640	0.2746
0.2521	3.22	6000	0.2797	0.2662
0.2654	3.43	6400	0.2762	0.2640
0.2586	3.64	6800	0.2642	0.2596
0.265	3.86	7200	0.2656	0.2794
0.2432	4.07	7600	0.2459	0.2497
0.226	4.29	8000	0.2533	0.2509
0.2385	4.5	8400	0.2332	0.2394
0.2332	4.72	8800	0.2500	0.2569
0.2358	4.93	9200	0.2384	0.2489
0.2169	5.14	9600	0.2410	0.2380
0.2038	5.36	10000	0.2426	0.2333
0.2109	5.57	10400	0.2480	0.2473
0.2147	5.79	10800	0.2341	0.2272
0.2153	6.0	11200	0.2402	0.2424
0.186	6.22	11600	0.2560	0.2489
0.1854	6.43	12000	0.2444	0.2402
0.1915	6.65	12400	0.2720	0.2531
0.1929	6.86	12800	0.2516	0.2342
0.1842	7.07	13200	0.2480	0.2304
0.1682	7.29	13600	0.2393	0.2276
0.1753	7.5	14000	0.2514	0.2263
0.1798	7.72	14400	0.2191	0.2178
0.1736	7.93	14800	0.2351	0.2197
0.1668	8.15	15200	0.2315	0.2194
0.1545	8.36	15600	0.2291	0.2079
0.1508	8.57	16000	0.2351	0.2134
0.1662	8.79	16400	0.2298	0.2197
0.1621	9.0	16800	0.2314	0.2219
0.1416	9.22	17200	0.2306	0.2192
0.1455	9.43	17600	0.2466	0.2184
0.1522	9.65	18000	0.2392	0.2255
0.1434	9.86	18400	0.2464	0.2208
0.1362	10.08	18800	0.2351	0.2095
0.127	10.29	19200	0.2373	0.2110
0.133	10.5	19600	0.2269	0.2031
0.1308	10.72	20000	0.2400	0.2096
0.1331	10.93	20400	0.2243	0.2083
0.125	11.15	20800	0.2334	0.2063
0.1236	11.36	21200	0.2195	0.2044
0.1263	11.58	21600	0.2263	0.2050
0.1235	11.79	22000	0.2217	0.2087
0.1301	12.0	22400	0.2332	0.2094
0.1123	12.22	22800	0.2195	0.2068
0.117	12.43	23200	0.2266	0.2110
0.1156	12.65	23600	0.2469	0.2063
0.1117	12.86	24000	0.2379	0.2035
0.1124	13.08	24400	0.2156	0.1963
0.106	13.29	24800	0.2310	0.1988
0.1066	13.5	25200	0.2334	0.1950
0.1069	13.72	25600	0.2230	0.2011
0.1089	13.93	26000	0.2233	0.2003
0.0977	14.15	26400	0.2273	0.1895
0.0972	14.36	26800	0.2265	0.1887
0.1005	14.58	27200	0.2196	0.1934
0.1058	14.79	27600	0.2213	0.1870
0.1027	15.01	28000	0.2361	0.1916
0.0886	15.22	28400	0.2275	0.1815
0.0885	15.43	28800	0.2230	0.1891
0.0911	15.65	29200	0.2237	0.1989
0.0923	15.86	29600	0.2200	0.1857
0.0868	16.08	30000	0.2248	0.1875
0.0812	16.29	30400	0.2240	0.1874
0.0829	16.51	30800	0.2198	0.1814
0.0832	16.72	31200	0.2328	0.1892
0.0822	16.93	31600	0.2283	0.1862
0.0828	17.15	32000	0.2283	0.1806
0.0791	17.36	32400	0.2197	0.1787
0.0801	17.58	32800	0.2249	0.1815
0.0804	17.79	33200	0.2304	0.1789
0.0833	18.01	33600	0.2235	0.1832
0.0762	18.22	34000	0.2358	0.1784
0.0688	18.44	34400	0.2183	0.1758
0.0751	18.65	34800	0.2169	0.1805
0.0729	18.86	35200	0.2296	0.1770
0.0681	19.08	35600	0.2380	0.1770
0.067	19.29	36000	0.2153	0.1777
0.0669	19.51	36400	0.2260	0.1742
0.0824	19.72	36800	0.0289	0.0310
0.0857	19.94	37200	0.0289	0.0322
0.0799	20.15	37600	0.0264	0.0298
0.0767	20.36	38000	0.0273	0.0318
0.079	20.58	38400	0.0274	0.0320
0.0791	20.79	38800	0.0279	0.0318
0.0805	21.01	39200	0.0285	0.0330
0.0622	21.22	39600	0.0263	0.0306
0.0622	21.44	40000	0.0290	0.0318
0.0672	21.65	40400	0.0278	0.0330
0.0706	21.86	40800	0.0270	0.0297
0.0619	22.08	41200	0.0288	0.0328
0.0633	22.29	41600	0.0256	0.0303
0.0618	22.51	42000	0.0263	0.0299
0.0576	22.72	42400	0.0273	0.0301
0.0583	22.94	42800	0.0282	0.0297
0.0565	23.15	43200	0.0256	0.0280
0.0557	23.37	43600	0.0268	0.0280
0.0548	23.58	44000	0.0266	0.0291
0.056	23.79	44400	0.0264	0.0290
0.0546	24.01	44800	0.0273	0.0284
0.0496	24.22	45200	0.0261	0.0279
0.0512	24.44	45600	0.0256	0.0281
0.0482	24.65	46000	0.0264	0.0285
0.0503	24.87	46400	0.0256	0.0268
0.0471	25.08	46800	0.0270	0.0282
0.0453	25.29	47200	0.0255	0.0267
0.0431	25.51	47600	0.0251	0.0264
0.0464	25.72	48000	0.0262	0.0261
0.0431	25.94	48400	0.0257	0.0265
0.0405	26.15	48800	0.0260	0.0251
0.0406	26.37	49200	0.0246	0.0250
0.0397	26.58	49600	0.0252	0.0254
0.0403	26.8	50000	0.0250	0.0256
0.0385	27.01	50400	0.0254	0.0241
0.0398	27.22	50800	0.0255	0.0242
0.0363	27.44	51200	0.0250	0.0236
0.0372	27.65	51600	0.0247	0.0232
0.0362	27.87	52000	0.0240	0.0226
0.0367	28.08	52400	0.0246	0.0224
0.0347	28.3	52800	0.0247	0.0229
0.0348	28.51	53200	0.0241	0.0229
0.0331	28.72	53600	0.0242	0.0224
0.0339	28.94	54000	0.0241	0.0220
0.0336	29.15	54400	0.0244	0.0221
0.0336	29.37	54800	0.0243	0.0215
0.0349	29.58	55200	0.0239	0.0217
0.0308	29.8	55600	0.0240	0.0219

Framework versions

Transformers 4.17.0
Pytorch 1.8.1+cu111
Datasets 2.2.1
Tokenizers 0.12.1

📄 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