How to tune alpha (lm_weight) and beta(word_count_weight) for our custom training set?

dbanka · December 29, 2017, 6:39am

Hi, I have trained using custom dataset of 1000 hours. I wanted to know how are the values of lm_weight, word_count_weight and valid_word_count_weight decided/optimized?

lissyx · January 5, 2018, 5:40am

I tend to recall that @reuben did that, he might be able to weight some knowledges

reuben · January 5, 2018, 9:52am

They are decided/optimized by doing a search over the possible values and choosing the value that yields the best WER on the validation set.

tanmayjain · April 15, 2020, 12:24pm

Hello,
Can you please share the file name, path and function name where we can tune this function?
Thanks

reuben · April 16, 2020, 2:54pm

github.com

mozilla/DeepSpeech/blob/master/lm_optimizer.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-
from __future__ import absolute_import, print_function

import absl.app
import optuna
import sys
import tensorflow.compat.v1 as tfv1

from deepspeech_training.evaluate import evaluate
from deepspeech_training.train import create_model
from deepspeech_training.util.config import Config, initialize_globals
from deepspeech_training.util.flags import create_flags, FLAGS
from deepspeech_training.util.logging import log_error
from deepspeech_training.util.evaluate_tools import wer_cer_batch
from ds_ctcdecoder import Scorer


def character_based():
    is_character_based = False

This file has been truncated. show original