Update README.md

README.md

@@ -26,20 +26,112 @@ model-index:
          value: 28.33
 ---
 
-# Wav2Vec2-Large-XLSR-53-Welsh
+# Wav2Vec2-Large-XLSR-Welsh
 
-Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Welsh using the [Common Voice](https://huggingface.co/datasets/common_voice) dataset.
+Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Welsh using the [Common Voice dataset](https://huggingface.co/datasets/common_voice).
 
-The code for training, using and evaluating this model can be found at GitHub: https://github.com/techiaith/xlsr-fine-tuning-week
+When using this model, make sure that your speech input is sampled at 16kHz.
 
-With a WER of 28.33%, here are some example predictions from the Common Voice Welsh test set:
 
-**Prediction:** rhedais i ffwrdd heb ddweud dim wrthi ym beth digwyddodd
+## Usage
 
-**Reference:** Rhedais i ffwrdd heb ddweud dim wrthi am beth ddigwyddodd.  
+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
 
-**Prediction:** ac yr oedd y ferch yn ofnus d
+test_dataset = load_dataset("common_voice", "cy", split="test[:2%]")
 
-**Reference:** Ac yr oedd y ferch yn ofnus.
+processor = Wav2Vec2Processor.from_pretrained("DewiBrynJones/wav2vec2-large-xlsr-welsh")
+model = Wav2Vec2ForCTC.from_pretrained("DewiBrynJones/wav2vec2-large-xlsr-welsh")
 
+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):
+	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["speech"][:2], 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 Welsh 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", "cy", split="test") 
+
+wer = load_metric("wer")
+
+processor = Wav2Vec2Processor.from_pretrained("DewiBrynJones/wav2vec2-large-xlsr-welsh")
+model = Wav2Vec2ForCTC.from_pretrained("DewiBrynJones/wav2vec2-large-xlsr-welsh")
+
+model.to("cuda")
+
+chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“]'  
+
+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):
+	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)
+
+# Preprocessing the datasets.
+# We need to read the aduio 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"])))
+```
+
+**Test Result**: 28.33% 
+
+
+# Training 
+
+A Docker based setup for training and evaluating this model can be found at GitHub: https://github.com/techiaith/xlsr-fine-tuning-week
+
+# Example Predictions
+
+| Prediction  | Reference  |
+|---|---|
+| rhedais i ffwrdd heb ddweud dim wrthi ym beth digwyddodd | Rhedais i ffwrdd heb ddweud dim wrthi am beth ddigwyddodd. |
+| ac yr oedd y ferch yn ofnus d | Ac yr oedd y ferch yn ofnus. |