wav2vec2-finetune-authentic-and-synthetic

arkitex/wav2vec2-finetune-authentic-and-synthetic

wav2vec2-finetune-authentic-and-synthetic is a automatic speech recognition model from arkitex. Use it when you need speech turned into text. It is set up for transformers. The card lists the license as apache-2.0.

should probably proofread and complete it, then remove this comment. --

Downloads · 30 days

13% of all-time downloads

All-time downloads

Public

Parameters

94.4M

34.4 GB on disk

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors378 MB · 100%

At a glance

Task: Automatic Speech Recognition
Library: transformers
License: apache-2.0
Model type: wav2vec2
Access: Public
Created: Apr 28, 2025
Updated: May 3, 2025
SHA: be91c4f8

Base models

facebook/wav2vec2-base

Datasets

common_voice_17_0

Model card

From the Hugging Face model README

wav2vec2-finetune-authentic-and-synthetic

This model is a fine-tuned version of facebook/wav2vec2-base on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

Loss: 0.4856
Wer: 0.3096

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0001
train_batch_size: 32
eval_batch_size: 8
seed: 42
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 1000
num_epochs: 3
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.0567	0.0548	500	2.9945	1.0
2.5069	0.1097	1000	1.4524	0.7744
0.94	0.1645	1500	1.0778	0.6329
0.7146	0.2194	2000	0.9130	0.5607
0.6139	0.2742	2500	0.9247	0.5226
0.5694	0.3291	3000	0.8565	0.5149
0.5277	0.3839	3500	0.8115	0.4786
0.5101	0.4387	4000	0.7730	0.4794
0.4919	0.4936	4500	0.7676	0.4485
0.4756	0.5484	5000	0.7997	0.4870
0.4657	0.6033	5500	0.7490	0.4536
0.4544	0.6581	6000	0.7065	0.4315
0.4412	0.7130	6500	0.7525	0.4413
0.4331	0.7678	7000	0.6989	0.4288
0.4332	0.8226	7500	0.6959	0.4296
0.4195	0.8775	8000	0.6629	0.4112
0.4068	0.9323	8500	0.6488	0.4083
0.3941	0.9872	9000	0.6760	0.4093
0.3784	1.0420	9500	0.6555	0.4001
0.3582	1.0969	10000	0.6495	0.4048
0.3577	1.1517	10500	0.6221	0.4014
0.345	1.2065	11000	0.6323	0.3862
0.3447	1.2614	11500	0.6067	0.3848
0.3383	1.3162	12000	0.6103	0.3843
0.3402	1.3711	12500	0.5990	0.3757
0.3256	1.4259	13000	0.6142	0.3870
0.327	1.4808	13500	0.6000	0.3741
0.3241	1.5356	14000	0.5835	0.3630
0.3206	1.5904	14500	0.5866	0.3698
0.3112	1.6453	15000	0.5728	0.3671
0.3128	1.7001	15500	0.5798	0.3628
0.3202	1.7550	16000	0.5766	0.3587
0.3098	1.8098	16500	0.5657	0.3569
0.3047	1.8646	17000	0.5769	0.3553
0.3035	1.9195	17500	0.5617	0.3596
0.2943	1.9743	18000	0.5521	0.3532
0.2808	2.0292	18500	0.5538	0.3469
0.2549	2.0840	19000	0.5402	0.3389
0.2562	2.1389	19500	0.5339	0.3333
0.2647	2.1937	20000	0.5278	0.3313
0.2508	2.2485	20500	0.5249	0.3296
0.2497	2.3034	21000	0.5310	0.3280
0.251	2.3582	21500	0.5288	0.3278
0.2468	2.4131	22000	0.5295	0.3298
0.241	2.4679	22500	0.5166	0.3235
0.2361	2.5228	23000	0.5179	0.3237
0.2389	2.5776	23500	0.5003	0.3165
0.2327	2.6324	24000	0.5108	0.3184
0.2384	2.6873	24500	0.4973	0.3139
0.2304	2.7421	25000	0.4945	0.3120
0.2306	2.7970	25500	0.4885	0.3130
0.2296	2.8518	26000	0.4906	0.3112
0.2264	2.9067	26500	0.4884	0.3085
0.2301	2.9615	27000	0.4856	0.3096

Framework versions

Transformers 4.51.3
Pytorch 2.6.0
Datasets 3.5.0
Tokenizers 0.21.1

Task: Automatic Speech Recognition
Library: transformers
Type: wav2vec2
License: apache-2.0
Created: Apr 28, 2025
Updated: May 3, 2025