Model reference · open weights

wav2vec2-large-english-TIMIT-phoneme

Audio speech31 · community Speech→text 1 build Open weights 1k dl/mo

wav2vec2-large-english-TIMIT-phoneme is an open-weight audio or speech model from speech31. wav2vec2-large-english-TIMIT-phoneme_v3 (BF16) weighs 1.3 GB; the smallest configuration that runs it is RTX 3060 12 GB.

Hugging Face ↗

What it is

Released by	speech31
Type	Audio & music
Task	Speech→text
Runs with	transformers
Released	2022-10-15
Popularity	1k downloads / month
Weights	1.3 GB (wav2vec2-large-english-TIMIT-phoneme_v3 (BF16), file size)
Licence	Open weights

What it runs on

Memory and cards for wav2vec2-large-english-TIMIT-phoneme_v3 (BF16)

Weights 1.3 GB (file size) · overhead about 1.6 GB.

Card	One stream	Counted memory
RTX 3060 12 GB	fits	11.6 GB
RTX 4060 Ti 16 GB	fits	15.4 GB
RTX 3090 24 GB	fits	23.4 GB
RTX 4090 24 GB	fits	23.4 GB
RTX 5090 32 GB	fits	31.0 GB
L40S 48 GB	fits	44.0 GB
A100 80 GB	fits	78.2 GB
H100 80 GB	fits	78.1 GB
RTX PRO 6000 Blackwell 96 GB	fits	93.8 GB
DGX Spark (GB10) 128 GB unified	fits	107 GB
H200 141 GB	fits	138 GB
B200 180 GB	fits	176 GB

Estimates, not measurements: the weights are the build's file size. A speech model's decoder keeps a small cache for every stream it transcribes, so memory grows with the streams and beams at once. Counted memory is 92 % of what CUDA reports for the card.

From the model card

What speech31 says about wav2vec2-large-english-TIMIT-phoneme

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

Loss: 0.3697
Cer: 0.0987

Read the full model card

Training and evaluation data

Training: TIMIT dataset training + validation set Evaluation: TIMIT dataset test set

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0003
train_batch_size: 32
eval_batch_size: 16
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 64
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 1000
num_epochs: 50
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per
2.2678	6.94	500	0.2347	0.0874
0.25	13.88	1000	0.3358	0.1122
0.2126	20.83	1500	0.3865	0.1131
0.1397	27.77	2000	0.4162	0.1085
0.0916	34.72	2500	0.4429	0.1086
0.0594	41.66	3000	0.3697	0.0987

Framework versions

Transformers 4.23.0.dev0
Pytorch 1.12.1.post201
Datasets 2.5.2.dev0
Tokenizers 0.12.1

Quoted from the model card on Hugging Face — the full card is behind the Hugging Face link above.