File size: 3,945 Bytes
19dfbde 32ca414 f52a598 19dfbde 32ca414 9ffa6b7 19dfbde 32ca414 d774a97 19dfbde 32ca414 9ffa6b7 32ca414 d774a97 32ca414 9ffa6b7 d774a97 9ffa6b7 d774a97 19dfbde 32ca414 19dfbde 32ca414 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 |
---
language:
- ha
license: apache-2.0
tags:
- generated_from_trainer
- robust-speech-event
- hf-asr-leaderboard
datasets:
- mozilla-foundation/common_voice_8_0
metrics:
- wer
base_model: facebook/wav2vec2-xls-r-300m
model-index:
- name: XLS-R-300M - Hausa
results:
- task:
type: automatic-speech-recognition
name: Speech Recognition
dataset:
name: Common Voice 8
type: mozilla-foundation/common_voice_8_0
args: ha
metrics:
- type: wer
value: 36.295
name: Test WER
- type: cer
value: 11.073
name: Test CER
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# XLS-R-300M - Hausa
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the common_voice dataset.
It achieves the following results on the evaluation set:
- Loss: 0.6094
- Wer: 0.5234
## 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: 16
- eval_batch_size: 8
- seed: 13
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine_with_restarts
- lr_scheduler_warmup_steps: 1000
- num_epochs: 100
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 2.9599 | 6.56 | 400 | 2.8650 | 1.0 |
| 2.7357 | 13.11 | 800 | 2.7377 | 0.9951 |
| 1.3012 | 19.67 | 1200 | 0.6686 | 0.7111 |
| 1.0454 | 26.23 | 1600 | 0.5686 | 0.6137 |
| 0.9069 | 32.79 | 2000 | 0.5576 | 0.5815 |
| 0.82 | 39.34 | 2400 | 0.5502 | 0.5591 |
| 0.7413 | 45.9 | 2800 | 0.5970 | 0.5586 |
| 0.6872 | 52.46 | 3200 | 0.5817 | 0.5428 |
| 0.634 | 59.02 | 3600 | 0.5636 | 0.5314 |
| 0.6022 | 65.57 | 4000 | 0.5780 | 0.5229 |
| 0.5705 | 72.13 | 4400 | 0.6036 | 0.5323 |
| 0.5408 | 78.69 | 4800 | 0.6119 | 0.5336 |
| 0.5225 | 85.25 | 5200 | 0.6105 | 0.5270 |
| 0.5265 | 91.8 | 5600 | 0.6034 | 0.5231 |
| 0.5154 | 98.36 | 6000 | 0.6094 | 0.5234 |
### Framework versions
- Transformers 4.16.1
- Pytorch 1.10.0+cu111
- Datasets 1.18.2
- Tokenizers 0.11.0
#### Evaluation Commands
1. To evaluate on `mozilla-foundation/common_voice_8_0` with split `test`
```bash
python eval.py --model_id anuragshas/wav2vec2-large-xls-r-300m-ha-cv8 --dataset mozilla-foundation/common_voice_8_0 --config ha --split test
```
### Inference With LM
```python
import torch
from datasets import load_dataset
from transformers import AutoModelForCTC, AutoProcessor
import torchaudio.functional as F
model_id = "anuragshas/wav2vec2-large-xls-r-300m-ha-cv8"
sample_iter = iter(load_dataset("mozilla-foundation/common_voice_8_0", "ha", split="test", streaming=True, use_auth_token=True))
sample = next(sample_iter)
resampled_audio = F.resample(torch.tensor(sample["audio"]["array"]), 48_000, 16_000).numpy()
model = AutoModelForCTC.from_pretrained(model_id)
processor = AutoProcessor.from_pretrained(model_id)
input_values = processor(resampled_audio, return_tensors="pt").input_values
with torch.no_grad():
logits = model(input_values).logits
transcription = processor.batch_decode(logits.numpy()).text
# => "kakin hade ya ke da kyautar"
```
### Eval results on Common Voice 8 "test" (WER):
| Without LM | With LM (run `./eval.py`) |
|---|---|
| 47.821 | 36.295 | |