Open-source wav2vec2-base-turkish-cv8 model - Free and accurate conversion of Turkish speech to text

Wav2vec2 Base Turkish Cv8

Developed by cahya

This is an automatic speech recognition (ASR) model fine-tuned on the Common Voice 8.0 Turkish dataset, capable of converting Turkish speech into text.

Speech Recognition

Transformers

Other#Turkish ASR #Common Voice fine-tuning #Low word error rate

Downloads 16

Release Time : 3/2/2022

Model Overview

This model is specifically designed for Turkish speech recognition tasks, achieving a 28.36% word error rate (WER) on the Common Voice evaluation set.

Model Features

Turkish Language Optimization

Specially optimized for Turkish language, performing well on Turkish speech recognition tasks

Based on Common Voice Data

Trained and evaluated using Mozilla Common Voice 8.0 Turkish dataset

Progressive Optimization

Continuously optimized during training, reducing word error rate from initial 27.52%

Model Capabilities

Turkish speech recognition

Speech-to-text

Use Cases

Speech Transcription

Turkish Speech Transcription

Convert Turkish speech content into text format

Word error rate 28.36%

Voice Assistant

Turkish Voice Command Recognition

Used for command recognition in Turkish voice assistant systems

🚀 Automatic Speech Recognition Model

This model is designed for automatic speech recognition, offering high - quality performance on Turkish speech data. It fine - tunes an existing model on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - TR dataset, achieving excellent results in evaluation.

🚀 Quick Start

This model is a fine - tuned version of [./checkpoint - 1000](https://huggingface.co/./checkpoint - 1000) on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - TR dataset. It achieves the following results on the evaluation set:

Loss: 0.3282
Wer: 0.2836

🔧 Technical Details

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0003
train_batch_size: 96
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 192
optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 100
num_epochs: 100.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0671	2.04	200	0.3079	0.2752
0.6433	4.08	400	0.2728	0.2848
0.5687	6.12	600	0.2882	0.3036
0.5355	8.16	800	0.2778	0.2920
0.5116	10.2	1000	0.2906	0.3014
0.5313	9.16	1200	0.2984	0.3273
0.4996	10.69	1400	0.3170	0.3344
0.4845	12.21	1600	0.3202	0.3634
0.5092	13.74	1800	0.3167	0.3373
0.4777	15.27	2000	0.3292	0.3386
0.4651	16.79	2200	0.3070	0.3427
0.461	18.32	2400	0.3149	0.3561
0.4481	19.85	2600	0.3292	0.3441
0.4479	21.37	2800	0.3142	0.3209
0.4305	22.9	3000	0.3525	0.3547
0.4254	24.43	3200	0.3414	0.3400
0.4066	25.95	3400	0.3118	0.3207
0.4043	27.48	3600	0.3418	0.3483
0.3985	29.01	3800	0.3254	0.3166
0.3982	30.53	4000	0.3306	0.3453
0.3929	32.06	4200	0.3262	0.3229
0.378	33.59	4400	0.3546	0.3336
0.4062	35.11	4600	0.3174	0.3457
0.3648	36.64	4800	0.3377	0.3357
0.3609	38.17	5000	0.3346	0.3520
0.3483	39.69	5200	0.3350	0.3526
0.3548	41.22	5400	0.3330	0.3406
0.3446	42.75	5600	0.3398	0.3372
0.3346	44.27	5800	0.3449	0.3288
0.3309	45.8	6000	0.3320	0.3144
0.326	47.33	6200	0.3400	0.3279
0.3189	48.85	6400	0.3400	0.3150
0.3165	50.38	6600	0.3359	0.2995
0.3132	51.91	6800	0.3343	0.3096
0.3092	53.44	7000	0.3224	0.3029
0.2995	54.96	7200	0.3205	0.2985
0.304	56.49	7400	0.3523	0.3034
0.2952	58.02	7600	0.3289	0.2934
0.2875	59.54	7800	0.3350	0.3008
0.2868	61.07	8000	0.3537	0.3227
0.2875	62.6	8200	0.3389	0.2970
0.2778	64.12	8400	0.3370	0.2960
0.2706	65.65	8600	0.3250	0.2802
0.2669	67.18	8800	0.3351	0.2903
0.2615	68.7	9000	0.3382	0.2989
0.2563	70.23	9200	0.3312	0.2975
0.2546	71.76	9400	0.3212	0.3003
0.2482	73.28	9600	0.3337	0.3091
0.2504	74.81	9800	0.3308	0.3110
0.2456	76.34	10000	0.3157	0.3118
0.2363	77.86	10200	0.3251	0.3144
0.2319	79.39	10400	0.3253	0.3038
0.2266	80.92	10600	0.3374	0.3038
0.2279	82.44	10800	0.3268	0.2964
0.2231	83.97	11000	0.3278	0.2950
0.2185	85.5	11200	0.3462	0.2981
0.2245	87.02	11400	0.3311	0.2895
0.223	88.55	11600	0.3325	0.2877
0.2121	90.08	11800	0.3337	0.2828
0.2126	91.6	12000	0.3325	0.2808
0.2027	93.13	12200	0.3277	0.2820
0.2058	94.66	12400	0.3308	0.2827
0.1991	96.18	12600	0.3279	0.2820
0.1991	97.71	12800	0.3300	0.2822
0.1986	99.24	13000	0.3285	0.2835

Framework versions

Transformers 4.17.0.dev0
Pytorch 1.10.2+cu102
Datasets 1.18.3
Tokenizers 0.11.0

Property	Details
Model Type	Fine - tuned model for automatic speech recognition
Training Data	MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - TR dataset

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