Update README.md

README.md

@@ -1,45 +1,123 @@
 ---
 language: sv
+datasets:
+- common_voice
 tags:
-- speech
 - audio
 - automatic-speech-recognition
+- speech
+- xlsr-fine-tuning-week
+license: apache-2.0
+model-index:
+- name: XLSR Wav2Vec2 Swedish by Marma
+  results:
+  - task: 
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    dataset:
+      name: Common Voice sv-SE
+      type: common_voice
+      args: sv  
+    metrics:
+       - name: Test WER
+         type: wer
+         value: 23.33
 ---
-## Wav2Vec 2.0 XLSR Swedish
 
-Swedish version of Wav2Vec2.0 XLSR finetuned on NST Swedish Dictation and evaluated using Common Voice
+# Wav2Vec2-Large-XLSR-53-Swedish
+
+Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Swedish using the [Common Voice](https://huggingface.co/datasets/common_voice)
+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
+
+test_dataset = load_dataset("common_voice", "sv-SE", split="test[:2%]").
+
+processor = Wav2Vec2Processor.from_pretrained("marma/wav2vec2-large-xlsr-swedish")
+model = Wav2Vec2ForCTC.from_pretrained("marma/wav2vec2-large-xlsr-swedish")
+
+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)
 
-**WER**: 23.3%
+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 Swedish test data of Common Voice.
 
-Does not work in the browser for some reason, but can be used as follows (code somewhat copied from Huggingface):
 
 ```python
-#!/usr/bin/env python3
-  
-from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
-import soundfile as sf
-from sys import argv,exit
 import torch
-import transformers
-from os.path import basename
+import torchaudio
+from datasets import load_dataset, load_metric
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+import re
+
+test_dataset = load_dataset("common_voice", "sv-SE", split="test")
+wer = load_metric("wer")
+
+processor = Wav2Vec2Processor.from_pretrained("marma/wav2vec2-large-xlsr-swedish")
+model = Wav2Vec2ForCTC.from_pretrained("marma/wav2vec2-large-xlsr-swedish")
+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
 
-if __name__ == '__main__':
-    if len(argv) < 3:
-        print(f'usage: {argv[0]} <model> <file 1>')
-        exit(1)
+result = test_dataset.map(evaluate, batched=True, batch_size=8)
 
-    device = "cpu"
+print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
+```
 
-    processor = Wav2Vec2Processor.from_pretrained(argv[1])
-    model = Wav2Vec2ForCTC.from_pretrained(argv[1]).to(device)
+**Test Result**: 23.33 %
 
-    f = argv[2]
-    s,sample_rate = sf.read(f)
-    input_values = processor(s, return_tensors="pt").to(device).input_values
-    logits = model(input_values).logits
-    predicted_ids = torch.argmax(logits, dim=-1)
 
-    transcription = processor.decode(predicted_ids[0])
+## Training
 
-    print(transcription)
-```
+The [NST Swedish Dictation](https://www.nb.no/sprakbanken/en/resource-catalogue/oai-nb-no-sbr-17/) was used for training.