wav2vec2-large-multilang-cv-ru Open Source Model - Freely Achieve Efficient Russian Speech Recognition

Wav2vec2 Large Multilang Cv Ru

Developed by cutten

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice dataset, primarily designed for Russian speech recognition tasks.

Speech Recognition

Transformers

Open Source License:Apache-2.0 #Russian speech recognition #Multilingual support #High-precision WER

Downloads 16

Release Time : 6/8/2022

Model Overview

This is a speech recognition model optimized for Russian, based on the wav2vec2 architecture and fine-tuned on the common_voice dataset.

Model Features

Multilingual pre-training foundation

Fine-tuned based on the facebook/wav2vec2-large-xlsr-53 model, which has been pre-trained on multiple languages

Russian optimization

Specifically optimized for Russian speech recognition tasks

Open-source license

Licensed under Apache 2.0, allowing commercial and research use

Model Capabilities

Russian speech recognition

Speech-to-text

Use Cases

Speech transcription

Russian speech transcription

Convert Russian speech content into text

Word Error Rate (WER) of 0.7037

Voice assistants

Russian voice command recognition

Used for voice command recognition in Russian voice assistant systems

🚀 wav2vec2-large-multilang-cv-ru

This model is a fine - tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice dataset, achieving certain results on the evaluation set.

🚀 Quick Start

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

Loss: 0.9734
Wer: 0.7037

📚 Documentation

Training and Evaluation Data

Although specific details about the training and evaluation data are not provided, this model is trained on the common_voice dataset.

Training Procedure

Training Hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0005
train_batch_size: 8
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e - 08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 1000
num_epochs: 30
mixed_precision_training: Native AMP

Training Results

Training Loss	Epoch	Step	Validation Loss	Wer
6.0328	0.79	500	3.0713	1.0
1.9426	1.58	1000	1.2048	0.9963
1.1285	2.37	1500	0.9825	0.9282
0.9462	3.15	2000	0.8836	0.8965
0.8274	3.94	2500	0.8134	0.8661
0.7106	4.73	3000	0.8033	0.8387
0.6545	5.52	3500	0.8309	0.8366
0.6013	6.31	4000	0.7667	0.8240
0.5599	7.1	4500	0.7740	0.8160
0.5027	7.89	5000	0.7796	0.8188
0.4588	8.68	5500	0.8204	0.7968
0.4448	9.46	6000	0.8277	0.7738
0.4122	10.25	6500	0.8292	0.7776
0.3816	11.04	7000	0.8548	0.7907
0.3587	11.83	7500	0.8245	0.7805
0.3374	12.62	8000	0.8371	0.7701
0.3214	13.41	8500	0.8311	0.7822
0.3072	14.2	9000	0.8940	0.7674
0.2929	14.98	9500	0.8788	0.7604
0.257	15.77	10000	0.8911	0.7633
0.2592	16.56	10500	0.8673	0.7604
0.2392	17.35	11000	0.9582	0.7810
0.232	18.14	11500	0.9340	0.7423
0.2252	18.93	12000	0.8874	0.7320
0.2079	19.72	12500	0.9436	0.7483
0.2003	20.5	13000	0.9573	0.7638
0.194	21.29	13500	0.9361	0.7308
0.188	22.08	14000	0.9704	0.7221
0.1754	22.87	14500	0.9668	0.7265
0.1688	23.66	15000	0.9680	0.7246
0.162	24.45	15500	0.9443	0.7066
0.1617	25.24	16000	0.9664	0.7265
0.1504	26.03	16500	0.9505	0.7189
0.1425	26.81	17000	0.9536	0.7112
0.134	27.6	17500	0.9674	0.7047
0.1301	28.39	18000	0.9852	0.7066
0.1314	29.18	18500	0.9766	0.7073
0.1219	29.97	19000	0.9734	0.7037

Framework Versions

Transformers 4.19.2
Pytorch 1.11.0
Datasets 2.2.2
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