model-facebookptbrlarge Open-source Speech Recognition Model - Free Deployment for Accurate Recognition of Brazilian Portuguese

Model Facebookptbrlarge

Developed by Vkt

A Brazilian Portuguese speech recognition model fine-tuned on the Common Voice dataset based on Facebook's wav2vec2-large-xlsr-53-portuguese model

Speech Recognition

Transformers

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

Downloads 22

Release Time : 6/7/2022

Model Overview

This is an automatic speech recognition (ASR) model optimized for Brazilian Portuguese, fine-tuned on the Common Voice dataset with a low word error rate

Model Features

High-accuracy speech recognition

Achieves a 13.22% word error rate (WER) on the Common Voice evaluation set

Optimized for Brazilian Portuguese

Specifically fine-tuned for Brazilian Portuguese speech characteristics

Based on wav2vec2 architecture

Utilizes Facebook's advanced wav2vec2-large-xlsr-53 architecture

Model Capabilities

Brazilian Portuguese speech-to-text

Continuous speech recognition

Audio content transcription

Use Cases

Speech transcription

Voice memo transcription

Convert Brazilian Portuguese voice memos to text

Highly accurate transcription results

Customer service conversation logs

Automatically record Brazilian Portuguese customer service conversations

Reduces manual transcription costs

Assistive technology

Voice input system

Provides voice input functionality for Brazilian Portuguese users

Enhances accessibility experience

🚀 model-facebookptbrlarge

This model is a fine - tuned version of facebook/wav2vec2-large-xlsr-53-portuguese on the common_voice dataset. It can achieve the following results on the evaluation set:

Loss: 0.2206
Wer: 0.1322

📚 Documentation

Training and evaluation data

Although specific details need to be further provided, this model is trained and evaluated on the common_voice dataset.

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
5.8975	0.29	400	0.4131	0.3336
0.5131	0.57	800	0.4103	0.3293
0.4846	0.86	1200	0.3493	0.3028
0.4174	1.14	1600	0.3055	0.2730
0.4105	1.43	2000	0.3283	0.3041
0.4028	1.72	2400	0.3539	0.3210
0.386	2.0	2800	0.2925	0.2690
0.3224	2.29	3200	0.2842	0.2665
0.3122	2.57	3600	0.2781	0.2472
0.3087	2.86	4000	0.2794	0.2692
0.2878	3.15	4400	0.2795	0.2537
0.2915	3.43	4800	0.2764	0.2478
0.2816	3.72	5200	0.2761	0.2366
0.283	4.0	5600	0.2641	0.2587
0.2448	4.29	6000	0.2489	0.2417
0.247	4.57	6400	0.2538	0.2422
0.25	4.86	6800	0.2660	0.2306
0.2256	5.15	7200	0.2477	0.2267
0.2225	5.43	7600	0.2364	0.2195
0.2217	5.72	8000	0.2319	0.2139
0.2272	6.0	8400	0.2489	0.2427
0.2016	6.29	8800	0.2404	0.2181
0.1973	6.58	9200	0.2532	0.2273
0.2101	6.86	9600	0.2590	0.2100
0.1946	7.15	10000	0.2414	0.2108
0.1845	7.43	10400	0.2485	0.2124
0.1861	7.72	10800	0.2405	0.2124
0.1851	8.01	11200	0.2449	0.2062
0.1587	8.29	11600	0.2510	0.2048
0.1694	8.58	12000	0.2290	0.2059
0.1637	8.86	12400	0.2376	0.2063
0.1594	9.15	12800	0.2307	0.1967
0.1537	9.44	13200	0.2274	0.2017
0.1498	9.72	13600	0.2322	0.2025
0.1516	10.01	14000	0.2323	0.1971
0.1336	10.29	14400	0.2249	0.1920
0.134	10.58	14800	0.2258	0.2055
0.138	10.86	15200	0.2250	0.1906
0.13	11.15	15600	0.2423	0.1920
0.1302	11.44	16000	0.2294	0.1849
0.1253	11.72	16400	0.2193	0.1889
0.1219	12.01	16800	0.2350	0.1869
0.1149	12.29	17200	0.2350	0.1903
0.1161	12.58	17600	0.2277	0.1899
0.1129	12.87	18000	0.2416	0.1855
0.1091	13.15	18400	0.2289	0.1815
0.1073	13.44	18800	0.2383	0.1799
0.1135	13.72	19200	0.2306	0.1819
0.1075	14.01	19600	0.2283	0.1742
0.0971	14.3	20000	0.2271	0.1851
0.0967	14.58	20400	0.2395	0.1809
0.1039	14.87	20800	0.2286	0.1808
0.0984	15.15	21200	0.2303	0.1821
0.0922	15.44	21600	0.2254	0.1745
0.0882	15.73	22000	0.2280	0.1836
0.0859	16.01	22400	0.2355	0.1779
0.0832	16.3	22800	0.2347	0.1740
0.0854	16.58	23200	0.2342	0.1739
0.0874	16.87	23600	0.2316	0.1719
0.0808	17.16	24000	0.2291	0.1730
0.0741	17.44	24400	0.2308	0.1674
0.0815	17.73	24800	0.2329	0.1655
0.0764	18.01	25200	0.2514	0.1711
0.0719	18.3	25600	0.2275	0.1578
0.0665	18.58	26000	0.2367	0.1614
0.0693	18.87	26400	0.2185	0.1593
0.0662	19.16	26800	0.2266	0.1678
0.0612	19.44	27200	0.2332	0.1602
0.0623	19.73	27600	0.2283	0.1670
0.0659	20.01	28000	0.2142	0.1626
0.0581	20.3	28400	0.2198	0.1646
0.063	20.59	28800	0.2251	0.1588
0.0618	20.87	29200	0.2186	0.1554
0.0549	21.16	29600	0.2251	0.1490
0.058	21.44	30000	0.2366	0.1559
0.0543	21.73	30400	0.2262	0.1535
0.0529	22.02	30800	0.2358	0.1519
0.053	22.3	31200	0.2198	0.1513
0.0552	22.59	31600	0.2234	0.1503
0.0492	22.87	32000	0.2191	0.1516
0.0488	23.16	32400	0.2321	0.1500
0.0479	23.45	32800	0.2152	0.1420
0.0453	23.73	33200	0.2202	0.1453
0.0485	24.02	33600	0.2235	0.1468
0.0451	24.3	34000	0.2192	0.1455
0.041	24.59	34400	0.2138	0.1438
0.0435	24.87	34800	0.2335	0.1423
0.0404	25.16	35200	0.2220	0.1409
0.0374	25.45	35600	0.2366	0.1437
0.0405	25.73	36000	0.2233	0.1428
0.0385	26.02	36400	0.2208	0.1414
0.0373	26.3	36800	0.2265	0.1420
0.0365	26.59	37200	0.2174	0.1402
0.037	26.88	37600	0.2249	0.1397
0.0379	27.16	38000	0.2173	0.1374
0.0354	27.45	38400	0.2212	0.1381
0.034	27.73	38800	0.2313	0.1364
0.0347	28.02	39200	0.2230	0.1356
0.0318	28.31	39600	0.2231	0.1357
0.0305	28.59	40000	0.2281	0.1366
0.0307	28.88	40400	0.2259	0.1342
0.0315	29.16	40800	0.2252	0.1332
0.0314	29.45	41200	0.2218	0.1328
0.0307	29.74	41600	0.2206	0.1322

Framework versions

Transformers 4.17.0
Pytorch 1.8.1+cu111
Datasets 2.2.1
Tokenizers 0.12.1

📄 License

This model is released 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