metadata

language:
  - he
license: apache-2.0
tags:
  - generated_from_trainer
  - he
  - robust-speech-event
datasets:
  - imvladikon/hebrew_speech_kan
  - imvladikon/hebrew_speech_coursera
metrics:
  - wer
base_model: facebook/wav2vec2-xls-r-300m
model-index:
  - name: wav2vec2-xls-r-300m-lm-hebrew
    results: []

wav2vec2-xls-r-300m-lm-hebrew

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset with adding ngram models according to Boosting Wav2Vec2 with n-grams in 🤗 Transformers

Usage

check package: https://github.com/imvladikon/wav2vec2-hebrew

or use transformers pipeline:

import torch
from datasets import load_dataset
from transformers import AutoModelForCTC, AutoProcessor
import torchaudio.functional as F


model_id = "imvladikon/wav2vec2-xls-r-300m-lm-hebrew"

sample_iter = iter(load_dataset("google/fleurs", "he_il", split="test", streaming=True))

sample = next(sample_iter)
resampled_audio = F.resample(torch.tensor(sample["audio"]["array"]), sample["audio"]["sampling_rate"], 16_000).numpy()

model = AutoModelForCTC.from_pretrained(model_id)
processor = AutoProcessor.from_pretrained(model_id)

input_values = processor(resampled_audio, return_tensors="pt").input_values

with torch.no_grad():
    logits = model(input_values).logits

transcription = processor.batch_decode(logits.numpy()).text
print(transcription)

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0003
train_batch_size: 64
eval_batch_size: 16
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 128
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 100
mixed_precision_training: Native AMP

Training results

Framework versions

Transformers 4.16.0.dev0
Pytorch 1.10.1+cu102
Datasets 1.17.1.dev0
Tokenizers 0.11.0