What is inverse text normalization?

By the Potok teamUpdated ~1 min read

What is inverse text normalization? It is the step that turns spoken forms into the written forms people expect, for example "twenty-five" into 25 or "three thirty p m" into 3:30 PM. Recognisers first write words, and this conversion then adds digits and symbols.

Transcribe a recording

Send your voice message, audio or video to @potok_now_bot on Telegram and the text comes back in the chat.

Open @potok_now_bot in Telegram

Files up to 2 GB · export to TXT, MD, DOCX, PDF, SRT and VTT · speaker labels and timestamps · about 30 languages

Key facts

  • Ordinary text normalization goes the other way, expanding digits and abbreviations into words so that a voice can read them.
  • Typical targets are numbers, times, dates, measurements, percentages and abbreviations.
  • Rules depend on the language and the style guide, so one region writes a date or decimal one way and another differently.
  • Ambiguity is the main risk: the same sound can be a word or a number, such as "one" in "the one I like".

What it looks like

spoken form                      written form
------------------------------   ----------------
twenty five                      25
three thirty p m                 3:30 PM
ten percent                      10%
two hundred meters              200 m
doctor smith                     Dr. Smith

More

See the full list of languages, the tools and the glossary.

Frequently asked questions

Why does the transcript sometimes show words instead of digits?

The conversion is a separate step, and tools differ in how far they apply it. Some leave numbers as words, particularly small ones and in unclear cases.

What is the opposite of inverse text normalization?

Text normalization. It expands written forms, such as 25 or Dr., into spoken words, which is what a speech synthesis system needs before reading text aloud.

Where do errors usually occur?

Phone numbers, long digit strings, addresses and units are common trouble spots. Always check figures and names in a transcript against the audio.

Does the style of numbers differ by country?

Yes. Thousands separators, decimal marks and date order vary, so a correct conversion in one region can look wrong in another. Settle on a style before editing.

How do I fix numbers in a Potok bot transcript?

Export it as DOCX or TXT, search for digits and spelled-out numbers, and compare each one with the recording. Do the same for times and measurements.

Sources

Each source was opened by hand on the date shown. Pages change, so open the source if a detail matters to you.

  1. Speech recognition (Wikipedia), Wikipedia · checked
  2. Transcription (linguistics) (Wikipedia), Wikipedia · checked

Try it on your own recording

Send your voice message, audio or video to @potok_now_bot on Telegram and the text comes back in the chat.

Open @potok_now_bot in Telegram