File size: 4,518 Bytes
d47bd37 67d044b d47bd37 67d044b d47bd37 67d044b d47bd37 a032e29 e69a778 a032e29 e69a778 d47bd37 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 |
---
language: lg
datasets:
- common_voice
metrics:
- wer
tags:
- audio
- automatic-speech-recognition
- speech
license: apache-2.0
model-index:
- name: Wav2Vec2 Luganda by Indonesian-NLP
results:
- task:
name: Speech Recognition
type: automatic-speech-recognition
dataset:
name: Common Voice lg
type: common_voice
args: lg
metrics:
- name: Test WER
type: wer
value: 7.53
---
# Automatic Speech Recognition for Luganda
This is the model built for the
[Mozilla Luganda Automatic Speech Recognition competition](https://zindi.africa/competitions/mozilla-luganda-automatic-speech-recognition).
It is a fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53)
model on the [Luganda Common Voice dataset](https://huggingface.co/datasets/common_voice) version 7.0.
We also provide a [live demo](https://huggingface.co/spaces/indonesian-nlp/luganda-asr) to test the model.
When using this model, make sure that your speech input is sampled at 16kHz.
## Usage
The model can be used directly (without a language model) as follows:
```python
import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
test_dataset = load_dataset("common_voice", "lg", split="test[:2%]")
processor = Wav2Vec2Processor.from_pretrained("indonesian-nlp/wav2vec2-luganda")
model = Wav2Vec2ForCTC.from_pretrained("indonesian-nlp/wav2vec2-luganda")
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the aduio files as arrays
def speech_file_to_array_fn(batch):
if "audio" in batch:
speech_array = torch.tensor(batch["audio"]["array"])
else:
speech_array, sampling_rate = torchaudio.load(batch["path"])
batch["speech"] = resampler(speech_array).squeeze().numpy()
return batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
inputs = processor(test_dataset[:2]["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
with torch.no_grad():
logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
predicted_ids = torch.argmax(logits, dim=-1)
print("Prediction:", processor.batch_decode(predicted_ids))
print("Reference:", test_dataset[:2]["sentence"])
```
## Evaluation
The model can be evaluated as follows on the Indonesian test data of Common Voice.
```python
import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re
test_dataset = load_dataset("common_voice", "lg", split="test")
wer = load_metric("wer")
processor = Wav2Vec2Processor.from_pretrained("indonesian-nlp/wav2vec2-luganda")
model = Wav2Vec2ForCTC.from_pretrained("indonesian-nlp/wav2vec2-luganda")
model.to("cuda")
chars_to_ignore = [",", "?", ".", "!", "-", ";", ":", '""', "%", "'", '"', "�", "‘", "’", "’"]
chars_to_ignore_regex = f'[{"".join(chars_to_ignore)}]'
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the audio files as arrays
def speech_file_to_array_fn(batch):
batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
if "audio" in batch:
speech_array = torch.tensor(batch["audio"]["array"])
else:
speech_array, sampling_rate = torchaudio.load(batch["path"])
batch["speech"] = resampler(speech_array).squeeze().numpy()
return batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
# Preprocessing the datasets.
# We need to read the audio files as arrays
def evaluate(batch):
inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
with torch.no_grad():
logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
pred_ids = torch.argmax(logits, dim=-1)
batch["pred_strings"] = processor.batch_decode(pred_ids)
return batch
result = test_dataset.map(evaluate, batched=True, batch_size=8)
print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
```
WER without KenLM: 15.38 %
WER With KenLM:
**Test Result**: 7.53 %
## Training
The Common Voice `train`, `validation`, and ... datasets were used for training as well as ... and ... # TODO
The script used for training can be found [here](https://github.com/indonesian-nlp/luganda-asr)
|