Update README.md
Browse files
README.md
CHANGED
@@ -23,12 +23,12 @@ model-index:
|
|
23 |
metrics:
|
24 |
- name: Test WER
|
25 |
type: wer
|
26 |
-
value:
|
27 |
---
|
28 |
|
29 |
# Wav2Vec2-Large-XLSR-53-rw
|
30 |
|
31 |
-
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
|
32 |
When using this model, make sure that your speech input is sampled at 16kHz.
|
33 |
|
34 |
## Usage
|
@@ -68,6 +68,8 @@ print("Prediction:", processor.batch_decode(predicted_ids))
|
|
68 |
print("Reference:", test_dataset["sentence"][:2])
|
69 |
```
|
70 |
|
|
|
|
|
71 |
|
72 |
## Evaluation
|
73 |
|
@@ -81,6 +83,7 @@ import torchaudio
|
|
81 |
from datasets import load_dataset, load_metric
|
82 |
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
|
83 |
import re
|
|
|
84 |
|
85 |
test_dataset = load_dataset("common_voice", "rw", split="test")
|
86 |
wer = load_metric("wer")
|
@@ -89,18 +92,30 @@ processor = Wav2Vec2Processor.from_pretrained("lucio/wav2vec2-large-xlsr-kinyarw
|
|
89 |
model = Wav2Vec2ForCTC.from_pretrained("lucio/wav2vec2-large-xlsr-kinyarwanda")
|
90 |
model.to("cuda")
|
91 |
|
92 |
-
chars_to_ignore_regex = '[
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
93 |
resampler = torchaudio.transforms.Resample(48_000, 16_000)
|
94 |
|
95 |
-
# Preprocessing the datasets.
|
96 |
-
# We need to read the audio files as arrays
|
97 |
def speech_file_to_array_fn(batch):
|
98 |
-
batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
|
99 |
speech_array, sampling_rate = torchaudio.load(batch["path"])
|
100 |
batch["speech"] = resampler(speech_array).squeeze().numpy()
|
|
|
101 |
return batch
|
102 |
|
103 |
-
|
|
|
|
|
|
|
|
|
|
|
104 |
|
105 |
# Preprocessing the datasets.
|
106 |
# We need to read the audio files as arrays
|
@@ -134,10 +149,11 @@ def chunked_wer(targets, predictions, chunk_size=None):
|
|
134 |
print("WER: {:2f}".format(100 * chunked_wer(result["sentence"], result["pred_strings"], chunk_size=4000)))
|
135 |
```
|
136 |
|
137 |
-
**Test Result**:
|
|
|
138 |
|
139 |
## Training
|
140 |
|
141 |
-
|
142 |
|
143 |
-
The script used for training
|
|
|
23 |
metrics:
|
24 |
- name: Test WER
|
25 |
type: wer
|
26 |
+
value: 40.59
|
27 |
---
|
28 |
|
29 |
# Wav2Vec2-Large-XLSR-53-rw
|
30 |
|
31 |
+
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 about 20% of the training data (limited to utterances without downvotes and shorter with 9.5 seconds), and validated on 2048 utterances from the validation set.
|
32 |
When using this model, make sure that your speech input is sampled at 16kHz.
|
33 |
|
34 |
## Usage
|
|
|
68 |
print("Reference:", test_dataset["sentence"][:2])
|
69 |
```
|
70 |
|
71 |
+
Prediction: ['yaherukaga gukora igitaramo y iki mu jyiwa na mul mumbiliki', 'ini rero ntibizashoboka ka nibo nkunrabibzi']
|
72 |
+
Reference: ['Yaherukaga gukora igitaramo nk’iki mu Mujyi wa Namur mu Bubiligi.', 'Ibi rero, ntibizashoboka, kandi nawe arabizi.']
|
73 |
|
74 |
## Evaluation
|
75 |
|
|
|
83 |
from datasets import load_dataset, load_metric
|
84 |
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
|
85 |
import re
|
86 |
+
import unidecode
|
87 |
|
88 |
test_dataset = load_dataset("common_voice", "rw", split="test")
|
89 |
wer = load_metric("wer")
|
|
|
92 |
model = Wav2Vec2ForCTC.from_pretrained("lucio/wav2vec2-large-xlsr-kinyarwanda")
|
93 |
model.to("cuda")
|
94 |
|
95 |
+
chars_to_ignore_regex = r'[!"#$%&()*+,./:;<=>?@\[\]\\_{}|~£¤¨©ª«¬®¯°·¸»¼½¾ðʺ˜˝ˮ‐–—―‚“”„‟•…″‽₋€™−√�]'
|
96 |
+
|
97 |
+
def remove_special_characters(batch):
|
98 |
+
batch["text"] = re.sub(r'[ʻʽʼ‘’´`]', r"'", batch["sentence"]) # normalize apostrophes
|
99 |
+
batch["text"] = re.sub(chars_to_ignore_regex, "", batch["text"]).lower().strip() # remove all other punctuation
|
100 |
+
batch["text"] = re.sub(r"(-| ?' ?| +)", " ", batch["text"]) # treat dash and apostrophe as word boundary
|
101 |
+
batch["text"] = unidecode.unidecode(batch["text"]) # strip accents
|
102 |
+
return batch
|
103 |
+
|
104 |
+
## Audio pre-processing
|
105 |
resampler = torchaudio.transforms.Resample(48_000, 16_000)
|
106 |
|
|
|
|
|
107 |
def speech_file_to_array_fn(batch):
|
|
|
108 |
speech_array, sampling_rate = torchaudio.load(batch["path"])
|
109 |
batch["speech"] = resampler(speech_array).squeeze().numpy()
|
110 |
+
batch["sampling_rate"] = 16_000
|
111 |
return batch
|
112 |
|
113 |
+
def cv_prepare(batch):
|
114 |
+
batch = remove_special_characters(batch)
|
115 |
+
batch = speech_file_to_array_fn(batch)
|
116 |
+
return batch
|
117 |
+
|
118 |
+
test_dataset = test_dataset.map(cv_prepare)
|
119 |
|
120 |
# Preprocessing the datasets.
|
121 |
# We need to read the audio files as arrays
|
|
|
149 |
print("WER: {:2f}".format(100 * chunked_wer(result["sentence"], result["pred_strings"], chunk_size=4000)))
|
150 |
```
|
151 |
|
152 |
+
**Test Result**: 40.59 %
|
153 |
+
|
154 |
|
155 |
## Training
|
156 |
|
157 |
+
Blocks of examples from the Common Voice training dataset (totaling about 100k examples, 20% of the available data) were used for training for 30k global steps, on 1 V100 GPU provided by OVHcloud. For validation, 2048 examples of the validation dataset were used.
|
158 |
|
159 |
+
The [script used for training](https://github.com/serapio/transformers/blob/feature/xlsr-finetune/examples/research_projects/wav2vec2/run_common_voice.py) is adapted from the [example script provided in the transformers repo](https://github.com/huggingface/transformers/blob/master/examples/research_projects/wav2vec2/run_common_voice.py).
|