Wav2vec2 Large Xlsr 53 Hebrew

🚀 wav2vec2-large-xlsr-53-hebrew

This model is a fine - tuned version of facebook/wav2vec2-large-xlsr-53 on several downloaded YouTube samples. It is designed for automatic speech recognition of Hebrew.

🚀 Quick Start

When using this model, make sure that your speech input is sampled at 16kHz.

✨ Features

• Language: Hebrew
• Datasets: Trained on samples from Common Voice and several downloaded YouTube samples.
• Metrics: Evaluated using Word Error Rate (WER).

📦 Installation

No specific installation steps are provided in the original README. However, you need to have the necessary Python libraries such as torch, torchaudio, datasets, and transformers installed. You can install them using pip:

pip install torch torchaudio datasets transformers

💻 Usage Examples

Basic Usage

import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
test_dataset = load_dataset("common_voice", "he", split="test[:2%]") # there is no common dataset for Hebrew, please, paste your data
processor = Wav2Vec2Processor.from_pretrained("imvladikon/wav2vec2-large-xlsr-53-hebrew")
model = Wav2Vec2ForCTC.from_pretrained("imvladikon/wav2vec2-large-xlsr-53-hebrew")
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])

Advanced Usage

import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re
test_dataset = load_dataset("common_voice", "he", split="test") # there is no common dataset for Hebrew, please, paste your data
wer = load_metric("wer")
processor = Wav2Vec2Processor.from_pretrained("imvladikon/wav2vec2-large-xlsr-53-hebrew")
model = Wav2Vec2ForCTC.from_pretrained("imvladikon/wav2vec2-large-xlsr-53-hebrew").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"])))

📄 License

This model is licensed under the Apache 2.0 license.