File size: 4,959 Bytes
f1ff9aa 032c0ae f1ff9aa 032c0ae f1ff9aa 032c0ae f1ff9aa 032c0ae 64ae049 f1ff9aa 64ae049 f1ff9aa |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 |
---
language: te
datasets:
- openslr
metrics:
- wer
tags:
- audio
- automatic-speech-recognition
- speech
- xlsr-fine-tuning-week
license: apache-2.0
model-index:
- name: Anurag Singh XLSR Wav2Vec2 Large 53 Telugu
results:
- task:
name: Speech Recognition
type: automatic-speech-recognition
dataset:
name: OpenSLR te
type: openslr
args: te
metrics:
- name: Test WER
type: wer
value: 44.98
---
# Wav2Vec2-Large-XLSR-53-Telugu
Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Telugu using the [OpenSLR SLR66](http://openslr.org/66/) dataset.
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
import pandas as pd
# Evaluation notebook contains the procedure to download the data
df = pd.read_csv("/content/te/test.tsv", sep="\\\\t")
df["path"] = "/content/te/clips/" + df["path"]
test_dataset = Dataset.from_pandas(df)
processor = Wav2Vec2Processor.from_pretrained("anuragshas/wav2vec2-large-xlsr-53-telugu")
model = Wav2Vec2ForCTC.from_pretrained("anuragshas/wav2vec2-large-xlsr-53-telugu")
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
```python
import torch
import torchaudio
from datasets import Dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re
from sklearn.model_selection import train_test_split
import pandas as pd
# Evaluation notebook contains the procedure to download the data
df = pd.read_csv("/content/te/test.tsv", sep="\\\\t")
df["path"] = "/content/te/clips/" + df["path"]
test_dataset = Dataset.from_pandas(df)
wer = load_metric("wer")
processor = Wav2Vec2Processor.from_pretrained("anuragshas/wav2vec2-large-xlsr-53-telugu")
model = Wav2Vec2ForCTC.from_pretrained("anuragshas/wav2vec2-large-xlsr-53-telugu")
model.to("cuda")
chars_to_ignore_regex = '[\\,\\?\\.\\!\\-\\_\\;\\:\\"\\“\\%\\‘\\”\\।\\’\'\\&]'
resampler = torchaudio.transforms.Resample(48_000, 16_000)
def normalizer(text):
# Use your custom normalizer
text = text.replace("\\\\\\
","\\
")
text = ' '.join(text.split())
text = re.sub(r'''([a-z]+)''','',text,flags=re.IGNORECASE)
text = re.sub(r'''%'''," శాతం ", text)
text = re.sub(r'''(/|-|_)'''," ", text)
text = re.sub("ై","ై", text)
text = text.strip()
return text
def speech_file_to_array_fn(batch):
batch["sentence"] = normalizer(batch["sentence"])
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**: 44.98%
## Training
70% of the OpenSLR Telugu dataset was used for training.
Train Split of annotations is [here](https://www.dropbox.com/s/xqc0wtour7f9h4c/train.tsv)
Test Split of annotations is [here](https://www.dropbox.com/s/qw1uy63oj4qdiu4/test.tsv)
Training Data Preparation notebook can be found [here](https://colab.research.google.com/drive/1_VR1QtY9qoiabyXBdJcOI29-xIKGdIzU?usp=sharing)
Training notebook can be found[here](https://colab.research.google.com/drive/14N-j4m0Ng_oktPEBN5wiUhDDbyrKYt8I?usp=sharing)
Evaluation notebook is [here](https://colab.research.google.com/drive/1SLEvbTWBwecIRTNqpQ0fFTqmr1-7MnSI?usp=sharing) |