Automatic Speech Recognition
Transformers
PyTorch
TensorBoard
speech-encoder-decoder
Generated from Trainer
# Load model directly
from transformers import AutoTokenizer, AutoModelForSpeechSeq2Seq
tokenizer = AutoTokenizer.from_pretrained("speech-seq2seq/wav2vec2-2-gpt2-medium")
model = AutoModelForSpeechSeq2Seq.from_pretrained("speech-seq2seq/wav2vec2-2-gpt2-medium")Quick Links
YAML Metadata Error:"model-index[0].name" is not allowed to be empty
This model was trained from scratch on the librispeech_asr dataset. It achieves the following results on the evaluation set:
- Loss: 3.5264
- Wer: 1.7073
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: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 3.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 4.4032 | 0.28 | 500 | 4.6724 | 1.9406 |
| 4.6417 | 0.56 | 1000 | 4.7143 | 1.8874 |
| 4.5725 | 0.84 | 1500 | 4.6413 | 1.9451 |
| 4.0178 | 1.12 | 2000 | 4.5470 | 1.8861 |
| 3.9084 | 1.4 | 2500 | 4.4360 | 1.8881 |
| 3.9297 | 1.68 | 3000 | 4.2814 | 1.8652 |
| 3.707 | 1.96 | 3500 | 4.1035 | 1.8320 |
| 3.1373 | 2.24 | 4000 | 3.9557 | 1.7762 |
| 3.3152 | 2.52 | 4500 | 3.7737 | 1.7454 |
| 2.9501 | 2.8 | 5000 | 3.5264 | 1.7073 |
Framework versions
- Transformers 4.17.0.dev0
- Pytorch 1.10.2+cu113
- Datasets 1.18.3
- Tokenizers 0.11.0
- Downloads last month
- 45
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("automatic-speech-recognition", model="speech-seq2seq/wav2vec2-2-gpt2-medium")