Model reference · open weights

wav2vec_asr_swbd

Name: wav2vec_asr_swbd
Brand: AxForge

Available as managed deployment Audio itaihay · community Speech→text 1 variants 2k dl/mo

wav2vec_asr_swbd is an open-weight audio or speech model from itaihay. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.

Request deployment All audio & music

Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.

What it is

Released by	itaihay
Type	Audio & music
Task	Speech→text
Runs with	transformers
Released	2022-03-31
Popularity	2k downloads / month
Licence	Open weights

About

What wav2vec_asr_swbd is

This model is a fine-tuned version of facebook/wav2vec2-large-robust-ft-swbd-300h on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.3052
Wer: 0.5302

Read the full model card

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: 0.0004
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 20
total_train_batch_size: 80
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 20
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.5445	0.29	500	0.9114	0.6197
0.9397	0.58	1000	0.5057	0.5902
0.8557	0.86	1500	0.4465	0.6264
0.7716	1.15	2000	0.4182	0.5594
0.7659	1.44	2500	0.4111	0.7048
0.7406	1.73	3000	0.3927	0.5944
0.6857	2.02	3500	0.3852	0.7118
0.7113	2.31	4000	0.3775	0.5608
0.6804	2.59	4500	0.3885	0.5759
0.6654	2.88	5000	0.3703	0.7226
0.6569	3.17	5500	0.3688	0.5972
0.6335	3.46	6000	0.3661	0.7278
0.6309	3.75	6500	0.3579	0.6324
0.6231	4.03	7000	0.3620	0.5770
0.6171	4.32	7500	0.3640	0.5772
0.6191	4.61	8000	0.3553	0.6075
0.6142	4.9	8500	0.3543	0.6126
0.5905	5.19	9000	0.3601	0.6319
0.5846	5.48	9500	0.3429	0.7343
0.5874	5.76	10000	0.3429	0.5962
0.5768	6.05	10500	0.3381	0.7410
0.5783	6.34	11000	0.3391	0.5823
0.5835	6.63	11500	0.3447	0.5821
0.5817	6.92	12000	0.3314	0.6890
0.5459	7.2	12500	0.3363	0.5727
0.5575	7.49	13000	0.3363	0.7387
0.5505	7.78	13500	0.3368	0.5685
0.55	8.07	14000	0.3330	0.5587
0.5523	8.36	14500	0.3338	0.5484
0.5116	8.65	15000	0.3350	0.4351
0.5263	8.93	15500	0.3254	0.6235
0.5265	9.22	16000	0.3297	0.6207
0.5265	9.51	16500	0.3279	0.6143
0.5172	9.8	17000	0.3260	0.5800
0.5028	10.09	17500	0.3259	0.5774
0.5062	10.37	18000	0.3259	0.5552
0.5112	10.66	18500	0.3201	0.6625
0.5149	10.95	19000	0.3184	0.6865
0.4939	11.24	19500	0.3152	0.6116
0.5065	11.53	20000	0.3172	0.5246
0.5129	11.82	20500	0.3129	0.5908
0.4909	12.1	21000	0.3152	0.6075
0.4865	12.39	21500	0.3160	0.5037
0.4805	12.68	22000	0.3139	0.5458
0.4691	12.97	22500	0.3225	0.5815
0.4534	13.26	23000	0.3168	0.5614
0.4661	13.54	23500	0.3135	0.6053
0.4636	13.83	24000	0.3120	0.5142
0.4554	14.12	24500	0.3127	0.5552
0.4602	14.41	25000	0.3117	0.5562
0.4521	14.7	25500	0.3106	0.4995
0.4369	14.99	26000	0.3100	0.5663
0.4249	15.27	26500	0.3110	0.5262
0.4321	15.56	27000	0.3106	0.5183
0.4293	15.85	27500	0.3091	0.5311
0.4537	16.14	28000	0.3134	0.4986
0.4258	16.43	28500	0.3138	0.4487
0.4347	16.71	29000	0.3091	0.5011
0.4615	17.0	29500	0.3068	0.5616
0.4163	17.29	30000	0.3115	0.5426
0.4074	17.58	30500	0.3079	0.5341
0.4121	17.87	31000	0.3047	0.5619
0.4219	18.16	31500	0.3085	0.5051
0.4049	18.44	32000	0.3084	0.5116
0.4119	18.73	32500	0.3071	0.5028
0.4129	19.02	33000	0.3064	0.5030
0.4143	19.31	33500	0.3040	0.5086
0.4013	19.6	34000	0.3057	0.5271
0.4162	19.88	34500	0.3052	0.5302

Framework versions

Transformers 4.18.0
Pytorch 1.11.0+cu113
Datasets 1.18.4
Tokenizers 0.11.6

From the published model card. Full card on the HuggingFace links in the sidebar.

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys wav2vec-asr-swbd for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (wav2vec-asr-swbd below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -F model="wav2vec-asr-swbd" -F file=@audio.mp3

Create an account — your API key is available in the console. 3M free tokens every 30 days with every new account.

Get API access

Sizes, precisions & builds

wav2vec_asr_swbd

— · BF16

Weights ↗

Details