udpated readme

README.md

@@ -37,11 +37,44 @@ The model can be used directly (without a language model) as follows:
 ```python
 import torch
 import torchaudio
-from datasets import load_dataset
+from datasets import load_dataset, load_metric, Dataset
 from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
-
-## TODO: load from OpenSLR https://openslr.org/41/
-##
+from datasets.utils.download_manager import DownloadManager
+from pathlib import Path
+import pandas as pd
+
+def load_dataset_javanese():
+    urls = [
+        "https://www.openslr.org/resources/41/jv_id_female.zip",
+        "https://www.openslr.org/resources/41/jv_id_male.zip"
+    ]
+    dm = DownloadManager()
+    download_dirs = dm.download_and_extract(urls)
+    data_dirs = [ 
+        Path(download_dirs[0])/"jv_id_female/wavs",
+        Path(download_dirs[1])/"jv_id_male/wavs",
+    ]
+    filenames = [ 
+        Path(download_dirs[0])/"jv_id_female/line_index.tsv",
+        Path(download_dirs[1])/"jv_id_male/line_index.tsv",
+    ]
+    
+    dfs = []
+    dfs.append(pd.read_csv(filenames[0], sep='\t', names=["path", "sentence"]))
+    dfs.append(pd.read_csv(filenames[1], sep='\t', names=["path", "client_id", "sentence"]))
+    dfs[1] = dfs[1].drop(["client_id"], axis=1)
+    
+    for i, dir in enumerate(data_dirs):
+        dfs[i]["path"] = dfs[i].apply(lambda row: str(data_dirs[i]) + "/" + row + ".wav", axis=1)
+    df = pd.concat(dfs)
+    # df = df.sample(frac=1, random_state=1).reset_index(drop=True)
+    dataset = Dataset.from_pandas(df)
+    dataset = dataset.remove_columns('__index_level_0__')
+    
+    return dataset.train_test_split(test_size=0.1, seed=1)
+
+dataset = load_dataset_javanese()
+test_dataset = dataset['test']
 
 processor = Wav2Vec2Processor.from_pretrained("cahya/wav2vec2-large-xlsr-javanese")
 model = Wav2Vec2ForCTC.from_pretrained("cahya/wav2vec2-large-xlsr-javanese")
@@ -51,15 +84,15 @@ 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):
-\tspeech_array, sampling_rate = torchaudio.load(batch["path"])
-\tbatch["speech"] = resampler(speech_array).squeeze().numpy()
-\treturn 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():
-\tlogits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
+    logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
 
 predicted_ids = torch.argmax(logits, dim=-1)
 
@@ -75,11 +108,45 @@ The model can be evaluated as follows on the Indonesian test data of Common Voic
 ```python
 import torch
 import torchaudio
-from datasets import load_dataset, load_metric
+from datasets import load_dataset, load_metric, Dataset
 from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
 import re
+from pathlib import Path
+import pandas as pd
+
+def load_dataset_javanese():
+    urls = [
+        "https://www.openslr.org/resources/41/jv_id_female.zip",
+        "https://www.openslr.org/resources/41/jv_id_male.zip"
+    ]
+    dm = DownloadManager()
+    download_dirs = dm.download_and_extract(urls)
+    data_dirs = [
+        Path(download_dirs[0])/"jv_id_female/wavs",
+        Path(download_dirs[1])/"jv_id_male/wavs",
+    ]
+    filenames = [
+        Path(download_dirs[0])/"jv_id_female/line_index.tsv",
+        Path(download_dirs[1])/"jv_id_male/line_index.tsv",
+    ]
+
+    dfs = []
+    dfs.append(pd.read_csv(filenames[0], sep='\t', names=["path", "sentence"]))
+    dfs.append(pd.read_csv(filenames[1], sep='\t', names=["path", "client_id", "sentence"]))
+    dfs[1] = dfs[1].drop(["client_id"], axis=1)
+
+    for i, dir in enumerate(data_dirs):
+        dfs[i]["path"] = dfs[i].apply(lambda row: str(data_dirs[i]) + "/" + row + ".wav", axis=1)
+    df = pd.concat(dfs)
+    # df = df.sample(frac=1, random_state=1).reset_index(drop=True)
+    dataset = Dataset.from_pandas(df)
+    dataset = dataset.remove_columns('__index_level_0__')
+
+    return dataset.train_test_split(test_size=0.1, seed=1)
+
+dataset = load_dataset_javanese()
+test_dataset = dataset['test']
 
-## TODO: load from OpenSLR https://openslr.org/41/
 wer = load_metric("wer")
 
 processor = Wav2Vec2Processor.from_pretrained("cahya/wav2vec2-large-xlsr-javanese")
@@ -93,24 +160,24 @@ 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):
-\tbatch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
-\tspeech_array, sampling_rate = torchaudio.load(batch["path"])
-\tbatch["speech"] = resampler(speech_array).squeeze().numpy()
-\treturn 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):
-\tinputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
+    inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
 
-\twith torch.no_grad():
-\t\tlogits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
+    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)
-\tbatch["pred_strings"] = processor.batch_decode(pred_ids)
-\treturn batch
+    batch["pred_strings"] = processor.batch_decode(pred_ids)
+    return batch
 
 result = test_dataset.map(evaluate, batched=True, batch_size=8)
 
@@ -124,4 +191,3 @@ print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"],
 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/cahya-wirawan/indonesian-speech-recognition) 
-(will be available soon)