Update README.md

README.md

@@ -39,35 +39,63 @@ When using this model, make sure that your speech input is sampled at 16kHz.
 The model can be used directly (without a language model) as follows:
 
 ```python
+import re
 import torch
+import librosa
 import torchaudio
-from datasets import load_dataset
+from datasets import load_dataset, load_metric
 from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+import soundfile as sf
 
-test_dataset = load_dataset("ma_speech_corpus", split="test")
 
-processor = Wav2Vec2Processor.from_pretrained("othrif/wav2vec2-large-xlsr-arabic")
-model = Wav2Vec2ForCTC.from_pretrained("othrif/wav2vec2-large-xlsr-arabic")
+dataset = load_dataset("ma_speech_corpus", split="test")
 
-resampler = torchaudio.transforms.Resample(48_000, 16_000)
+processor = Wav2Vec2Processor.from_pretrained("othrif/wav2vec2-large-xlsr-moroccan")
+model = Wav2Vec2ForCTC.from_pretrained("othrif/wav2vec2-large-xlsr-moroccan")
+model.to("cuda")
+
+
+chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\'\�]'
+
+def remove_special_characters(batch):
+    batch["text"] = re.sub(chars_to_ignore_regex, "", batch["sentence"]).lower() + " "
+    return batch
+
+
+dataset = dataset.map(remove_special_characters)
+dataset = dataset.select(range(10))
 
-# Preprocessing the datasets.
-# We need to read the audio files as arrays
 def speech_file_to_array_fn(batch):
+    start, stop = batch['segment'].split('_')
     speech_array, sampling_rate = torchaudio.load(batch["path"])
-    batch["speech"] = resampler(speech_array).squeeze().numpy()
+    speech_array, sampling_rate = sf.read(batch["path"], start=int(float(start) * sampling_rate),
+                                          stop=int(float(stop) * sampling_rate))
+    batch["speech"] = librosa.resample(speech_array, sampling_rate, 16_000)
+    batch["sampling_rate"] = 16_000
+    batch["target_text"] = batch["text"]
     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
+dataset = dataset.map(
+    speech_file_to_array_fn
+)
+
+def predict(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["predicted"] = processor.batch_decode(pred_ids)
+    return batch
 
-predicted_ids = torch.argmax(logits, dim=-1)
+dataset = dataset.map(predict, batched=True, batch_size=32)
 
-print("Prediction:", processor.batch_decode(predicted_ids))
-print("Reference:", test_dataset["sentence"][:2])
+for reference, predicted in zip(dataset["sentence"], dataset["predicted"]):
+    print("reference:", reference)
+    print("predicted:", predicted)
+    print("--")
 ```
 
 
@@ -77,60 +105,67 @@ The model can be evaluated as follows on the Arabic test data of Common Voice.
 
 
 ```python
-import re
 import torch
-import librosa
 import torchaudio
 from datasets import load_dataset, load_metric
 from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
-import soundfile as sf
+import re
 
-eval_dataset = load_dataset("ma_speech_corpus", split="test")
+test_dataset = load_dataset("ma_speech_corpus", split="test") 
 wer = load_metric("wer")
 
-processor = Wav2Vec2Processor.from_pretrained("othrif/wav2vec2-large-xlsr-moroccan")
+processor = Wav2Vec2Processor.from_pretrained("othrif/wav2vec2-large-xlsr-moroccan") 
 model = Wav2Vec2ForCTC.from_pretrained("othrif/wav2vec2-large-xlsr-moroccan")
 model.to("cuda")
 
-chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\'\�]'
-
-def remove_special_characters(batch):
-    batch["text"] = re.sub(chars_to_ignore_regex, "", batch["sentence"]).lower() + " "
-    return batch
-
-
-eval_dataset = eval_dataset.map(remove_special_characters, remove_columns=["sentence"])
-#eval_dataset = eval_dataset.select(range(100))
+chars_to_ignore_regex = '[0\\\\,\\\\?\\\\.\\\\!\\\\-\\\\;\\\\:\\\\"\\\\“\\\\%\\\\‘\\\\”\\\\�\\
+\\\\@\\\\ـ\\\\؟\\\\*\\\\ \\\\#\\\\'\\\\ \\\\…\\\\\\\\u2003]'
+#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):
-    start, stop = batch['segment'].split('_')
+    batch["text"] = re.sub(chars_to_ignore_regex, '', batch["text"]).lower()
+    batch["text"] = re.sub('[a-zA-z]', '', batch["text"]).lower() + " "
+        batch["text"] = re.sub('[ًٌٍَُِ~]', '', batch["text"]).lower() + " "
+
+        # batch["text"] = re.sub('\\\\\\
+','', batch["text"])
+        batch["text"] = re.sub("[إأٱآا]", "ا", batch["text"])
+        batch["text"] = re.sub("ڸ", "ل", batch["text"])
+        noise = re.compile(""" ّ    | # Tashdid
+                             َ    | # Fatha
+                             ً    | # Tanwin Fath
+                             ُ    | # Damma
+                             ٌ    | # Tanwin Damm
+                             ِ    | # Kasra
+                             ٍ    | # Tanwin Kasr
+                             ْ    | # Sukun
+                             ـ     # Tatwil/Kashida
+                         """, re.VERBOSE)
+        batch["text"] = re.sub(noise, '', batch["text"])
+        batch["text"] = re.sub('ٖ', '', batch["text"]).lower() + " "
     speech_array, sampling_rate = torchaudio.load(batch["path"])
-    speech_array, sampling_rate = sf.read(batch["path"], start=int(float(start) * sampling_rate),
-                                          stop=int(float(stop) * sampling_rate))
-    batch["speech"] = librosa.resample(speech_array, sampling_rate, 16_000)
-    batch["sampling_rate"] = 16_000
-    batch["target_text"] = batch["text"]
+    batch["speech"] = resampler(speech_array).squeeze().numpy()
     return batch
 
+test_dataset = test_dataset.map(speech_file_to_array_fn)
 
-eval_dataset = eval_dataset.map(
-    speech_file_to_array_fn,
-    remove_columns=eval_dataset.column_names
-)
-
+# 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
+    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 = eval_dataset.map(evaluate, batched=True, batch_size=32)
+result = test_dataset.map(evaluate, batched=True, batch_size=8)
 
-print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["target_text"])))
+print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["text"])))
 ```
 
 **Test Result**: 66.45