File size: 5,892 Bytes
ab6470e b1be9a4 ab6470e b1be9a4 ab6470e b1be9a4 ab6470e b1be9a4 ab6470e b1be9a4 ab6470e b1be9a4 ab6470e b1be9a4 ab6470e b1be9a4 ab6470e b1be9a4 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 |
---
language: rw
datasets:
- common_voice
metrics:
- wer
tags:
- audio
- automatic-speech-recognition
- speech
- xlsr-fine-tuning-week
license: apache-2.0
model-index:
- name: XLSR Wav2Vec2 Large Kinyarwanda by Lucio
results:
- task:
name: Speech Recognition
type: automatic-speech-recognition
dataset:
name: Common Voice rw
type: common_voice
args: rw
metrics:
- name: Test WER
type: wer
value: 47.99
---
# Wav2Vec2-Large-XLSR-53-rw
Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Kinyarwanda using the [Common Voice](https://huggingface.co/datasets/common_voice) dataset, using the validation set for training, and taking 12% of the test data for validation.
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
# WARNING! This will download and extract to use about 80GB on disk.
test_dataset = load_dataset("common_voice", "rw", split="test[:2%]")
processor = Wav2Vec2Processor.from_pretrained("lucio/wav2vec2-large-xlsr-kinyarwanda")
model = Wav2Vec2ForCTC.from_pretrained("lucio/wav2vec2-large-xlsr-kinyarwanda")
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):
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["sentence"][:2])
```
## Evaluation
The model can be evaluated as follows on the Kinyarwanda test data of Common Voice. Note that to even load the test data, the whole 40GB Kinyarwanda dataset will be downloaded and extracted into another 40GB directory, so you will need that space available on disk (e.g. not possible in the free tier of Google Colab). This script uses the `chunked_wer` function from [pcuenq](https://huggingface.co/pcuenq/wav2vec2-large-xlsr-53-es).
```python
import jiwer
import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re
test_dataset = load_dataset("common_voice", "rw", split="test")
wer = load_metric("wer")
processor = Wav2Vec2Processor.from_pretrained("lucio/wav2vec2-large-xlsr-kinyarwanda")
model = Wav2Vec2ForCTC.from_pretrained("lucio/wav2vec2-large-xlsr-kinyarwanda")
model.to("cuda")
chars_to_ignore_regex = '[\\[\\],?.!;:%\\'"‘’“”(){}‟ˮ´ʺ″«»/…‽�–-]'
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()
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, remove_columns=['path'])
# 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)
def chunked_wer(targets, predictions, chunk_size=None):
if chunk_size is None: return jiwer.wer(targets, predictions)
start = 0
end = chunk_size
H, S, D, I = 0, 0, 0, 0
while start < len(targets):
chunk_metrics = jiwer.compute_measures(targets[start:end], predictions[start:end])
H = H + chunk_metrics["hits"]
S = S + chunk_metrics["substitutions"]
D = D + chunk_metrics["deletions"]
I = I + chunk_metrics["insertions"]
start += chunk_size
end += chunk_size
return float(S + D + I) / float(H + S + D)
print("WER: {:2f}".format(100 * chunked_wer(result["sentence"], result["pred_strings"], chunk_size=4000)))
```
**Test Result**: 47.99 %
## Training
The Common Voice `validation` dataset was used for training, with 12% of the test dataset used for validation, trained on 1 V100 GPU for 48 hours (20 epochs).
The script used for training was just the `run_finetuning.py` script provided in OVHcloud's `databuzzword/hf-wav2vec` image. |