Model reference · open weights

wav2vec2-speechdat

Name: wav2vec2-speechdat
Brand: AxForge

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

wav2vec2-speechdat is an open-weight audio or speech model from birgermoell. 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	birgermoell
Type	Audio & music
Task	Speech→text
Runs with	transformers
Released	2022-03-02
Popularity	1k downloads / month
Licence	Open weights

About

What wav2vec2-speechdat is

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

Loss: 0.4578
Wer: 0.2927

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: 0.0003
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: 15.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.01	100	3.6252	1.0
No log	0.02	200	3.1906	1.0
No log	0.03	300	3.1090	1.0
No log	0.04	400	1.8796	0.9955
6.2575	0.05	500	1.3515	0.9058
6.2575	0.06	600	1.1209	0.8328
6.2575	0.07	700	1.1404	0.8309
6.2575	0.09	800	1.0599	0.8021
6.2575	0.1	900	0.9901	0.8335
0.7737	0.11	1000	0.8846	0.7400
0.7737	0.12	1100	0.9971	0.7820
0.7737	0.13	1200	0.8665	0.7123
0.7737	0.14	1300	0.8490	0.7366
0.7737	0.15	1400	0.8250	0.6765
0.6183	0.16	1500	0.8291	0.6965
0.6183	0.17	1600	0.7946	0.6823
0.6183	0.18	1700	0.8239	0.6894
0.6183	0.19	1800	0.8282	0.6796
0.6183	0.2	1900	0.7645	0.6518
0.561	0.21	2000	0.7530	0.6367
0.561	0.22	2100	0.7296	0.6177
0.561	0.24	2200	0.7527	0.6498
0.561	0.25	2300	0.7210	0.6316
0.561	0.26	2400	0.7938	0.6757
0.5402	0.27	2500	0.7485	0.6372
0.5402	0.28	2600	0.7146	0.6133
0.5402	0.29	2700	0.7308	0.6626
0.5402	0.3	2800	0.7078	0.5949
0.5402	0.31	2900	0.7679	0.6373
0.5303	0.32	3000	0.7263	0.6502
0.5303	0.33	3100	0.6613	0.5846
0.5303	0.34	3200	0.6784	0.5783
0.5303	0.35	3300	0.6908	0.5833
0.5303	0.36	3400	0.6595	0.5826
0.503	0.37	3500	0.6717	0.5938
0.503	0.39	3600	0.6938	0.5791
0.503	0.4	3700	0.6677	0.6052
0.503	0.41	3800	0.6544	0.5554
0.503	0.42	3900	0.6514	0.5728
0.4959	0.43	4000	0.6847	0.6188
0.4959	0.44	4100	0.6626	0.5869
0.4959	0.45	4200	0.6670	0.5700
0.4959	0.46	4300	0.6596	0.5846
0.4959	0.47	4400	0.6523	0.5468
0.4824	0.48	4500	0.6392	0.5688
0.4824	0.49	4600	0.6561	0.5687
0.4824	0.5	4700	0.6697	0.5817
0.4824	0.51	4800	0.6348	0.5608
0.4824	0.52	4900	0.6561	0.5600
0.4714	0.54	5000	0.6522	0.6181
0.4714	0.55	5100	0.6858	0.5921
0.4714	0.56	5200	0.6706	0.5497
0.4714	0.57	5300	0.7123	0.5768
0.4714	0.58	5400	0.6599	0.6100
0.471	0.59	5500	0.6421	0.5626
0.471	0.6	5600	0.6395	0.5753
0.471	0.61	5700	0.6788	0.5481
0.471	0.62	5800	0.6386	0.5516
0.471	0.63	5900	0.6694	0.5913
0.4707	0.64	6000	0.6251	0.5699
0.4707	0.65	6100	0.6243	0.5567
0.4707	0.66	6200	0.6645	0.5629
0.4707	0.67	6300	0.6296	0.5895
0.4707	0.69	6400	0.6078	0.5183
0.4632	0.7	6500	0.6270	0.5619
0.4632	0.71	6600	0.6050	0.5336
0.4632	0.72	6700	0.6185	0.5449
0.4632	0.73	6800	0.6281	0.5645
0.4632	0.74	6900	0.5877	0.5084
0.4514	0.75	7000	0.6199	0.5403
0.4514	0.76	7100	0.6293	0.5275
0.4514	0.77	7200	0.6290	0.5447
0.4514	0.78	7300	0.6130	0.5373
0.4514	0.79	7400	0.6138	0.5285
0.4457	0.8	7500	0.6040	0.5259
0.4457	0.81	7600	0.6220	0.5686
0.4457	0.82	7700	0.5915	0.5164
0.4457	0.84	7800	0.6270	0.5289
0.4457	0.85	7900	0.6224	0.5515
0.4458	0.86	8000	0.6161	0.5323
0.4458	0.87	8100	0.5827

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 wav2vec2-speechdat for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (wav2vec2-speechdat 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="wav2vec2-speechdat" -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

wav2vec2-speechdat

— · BF16

Weights ↗

Details