The wav2vec2-large-xlsr-53-W2V2-TATAR-SMALL Open-Source Model - Accurately Achieve Automatic Speech Recognition for Tatar Language

Wav2vec2 Large Xlsr 53 W2V2 TATAR SMALL

Developed by emre

This model is a Tatar automatic speech recognition model fine-tuned on the Common Voice 8 dataset based on facebook/wav2vec2-large-xlsr-53, with a test set WER of 53.16%.

Speech Recognition

Transformers

OtherOpen Source License:Apache-2.0 #Tatar speech recognition #Low-resource language optimization #WER optimization model

Downloads 30

Release Time : 3/2/2022

Model Overview

This is an automatic speech recognition model optimized for Tatar, based on the wav2vec2 architecture and fine-tuned on the Common Voice dataset.

Model Features

Multilingual support

Based on the XLSR-53 pre-trained model with cross-lingual representation capabilities

Speech recognition optimization

Specially fine-tuned and optimized for Tatar language

Efficient training

Uses mixed-precision training and gradient accumulation techniques to improve training efficiency

Model Capabilities

Tatar speech-to-text

Speech recognition

Speech transcription

Use Cases

Speech transcription

Tatar speech transcription

Convert Tatar speech content into text

WER 53.16%

Voice assistant

Tatar voice command recognition

Speech recognition module for Tatar voice assistants or voice control systems

🚀 wav2vec2-large-xlsr-53-W2V2-TATAR-SMALL

This model is a fine - tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice dataset. It can be used for automatic speech recognition, achieving certain results on the evaluation set and contributing to the field of speech recognition.

📚 Documentation

Model Information

Property	Details
Model Type	Fine - tuned version of facebook/wav2vec2-large-xlsr-53
Training Data	common_voice
Base Model	facebook/wav2vec2-large-xlsr-53

Evaluation Results

This model achieves the following results on the evaluation set:

Loss: 0.4714
Wer: 0.5316

Model Index

Name: wav2vec2-large-xlsr-53-W2V2-TATAR-SMALL
Results:
- Task:
  - Type: automatic-speech-recognition
  - Name: Automatic Speech Recognition
- Dataset:
  - Name: Common Voice 8
  - Type: mozilla-foundation/common_voice_8_0
  - Args: tt
- Metrics:
  - Type: wer
  - Value: 53.16
  - Name: Test WER

🔧 Technical Details

Training Hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0003
train_batch_size: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
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: 500
num_epochs: 30
mixed_precision_training: Native AMP

Training Results

Training Loss	Epoch	Step	Validation Loss	Wer
6.2446	1.17	400	3.2621	1.0
1.739	2.35	800	0.5832	0.7688
0.4718	3.52	1200	0.4785	0.6824
0.3574	4.69	1600	0.4814	0.6792
0.2946	5.86	2000	0.4484	0.6506
0.2674	7.04	2400	0.4612	0.6225
0.2349	8.21	2800	0.4600	0.6050
0.2206	9.38	3200	0.4772	0.6048
0.2072	10.56	3600	0.4676	0.6106
0.1984	11.73	4000	0.4816	0.6079
0.1793	12.9	4400	0.4616	0.5836
0.172	14.08	4800	0.4808	0.5860
0.1624	15.25	5200	0.4854	0.5820
0.156	16.42	5600	0.4609	0.5656
0.1448	17.59	6000	0.4926	0.5817
0.1406	18.77	6400	0.4638	0.5654
0.1337	19.94	6800	0.4731	0.5652
0.1317	21.11	7200	0.4861	0.5639
0.1179	22.29	7600	0.4766	0.5521
0.1197	23.46	8000	0.4824	0.5584
0.1096	24.63	8400	0.5006	0.5559
0.1038	25.81	8800	0.4994	0.5440
0.0992	26.98	9200	0.4867	0.5405
0.0984	28.15	9600	0.4798	0.5361
0.0943	29.33	10000	0.4714	0.5316

Framework Versions

Transformers 4.11.3
Pytorch 1.10.0+cu111
Datasets 1.14.0
Tokenizers 0.10.3

📄 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