Numbers Spelling

[MobinaPoulaei]

Checks

• This template is only for research question, not usage problems, feature requests or bug reports.
• I have thoroughly reviewed the project documentation and read the related paper(s).
• I have searched for existing issues, including closed ones, no similar questions.
• I am using English to submit this issue to facilitate community communication.

Question details

Hi,

I tested the model with the sentence:

"I have 2345 apples."

The generated speech did not read the number correctly. It seems that the model does not properly handle numbers larger than 1000. Is this a known limitation, and is there a recommended way to address it?

Thanks.

[CHRONOMASTER]

Try replacing all multi digit numbers with the word equivalent.
For example, instead of 2345, replace it with two thousand three hundred forty-five.

[MobinaPoulaei]

Try replacing all multi digit numbers with the word equivalent. For example, instead of 2345, replace it with two thousand three hundred forty-five.

Yes, I know that would work, but I was curious whether there is a better approach than manually normalizing numbers before inference.

[celguar]

There is - let library like num2words automatically normalize numbers before inference

def convert_numbers_to_words(text):
    from num2words import num2words
    """Convert all numbers in text to their word representation"""

    def replace_number(match):
        num = match.group()
        words = num2words(num)
        # Remove all commas
        words = words.replace(',', '')
        return words

    # Find all numbers in text
    pattern = r'\b\d+\b'
    return re.sub(pattern, replace_number, text)

[zhu-han]

Text normalization is the solution. We mentioned this in README.md: For better results with Arabic numerals, normalize them to words first (e.g., "123" → "one hundred twenty-three") with text normalization tools (e.g., [WeTextProcessing](https://github.com/wenet-e2e/WeTextProcessing)).