THA: Weighted Finite-State Text Normalization and Inverse Text Normalization for Khmer

Seanghay Yath

cs.CLSep 25, 2026

THA: Weighted Finite-State Text Normalization and Inverse Text Normalization for Khmer

Authors: Seanghay Yath

Organizations: Digital Government Committee, Cambodia

Abstract

Text-to-speech needs written text in spoken form, and speech recognition output needs the reverse. For Khmer, neither direction has a maintained open-source tool, and the script makes both harder: words are not separated by spaces, and number words occur inside ordinary words. We present Tha, a Khmer text normalization and inverse text normalization toolkit built from weighted finite-state transducers. It segments and classifies a whole line in one shortest-path search, and a second transducer rejects token boundaries inside a Khmer syllable. On Google's Khmer test suite, Tha agrees with the reference on all 274 cardinals up to one spelling variant, and on 2,906 real TTS prompts, 153 of the 158 sentences it rewrites are correct. Tha is open source under the Apache 2.0 license.

Figures & tables

1	\khmerfont មួយ	10	\khmerfont ដប់	10 2	\khmerfont រយ
2	\khmerfont ពីរ	20	\khmerfont ម្ភៃ	10 3	\khmerfont ពាន់
3	\khmerfont បី	30	\khmerfont សាមសិប ( \khmerfont សាម)	10 4	\khmerfont ម៉ឺន
4	\khmerfont បួន	40	\khmerfont សែសិប ( \khmerfont សែ)	10 5	\khmerfont សែន
5	\khmerfont ប្រាំ	50	\khmerfont ហាសិប ( \khmerfont ហា)	10 6	\khmerfont លាន
6	\khmerfont ប្រាំមួយ	60	\khmerfont ហុកសិប ( \khmerfont ហុក)	10 9	\khmerfont ប៊ីលាន
7	\khmerfont ប្រាំពីរ	70	\khmerfont ចិតសិប ( \khmerfont ចិត)		\khmerfont ពាន់លាន

Table 1: Khmer number words. Short spoken forms of the tens are in parentheses; they are only used before a digit ( \khmerfont ហុកពីរ, 62).

Panel of Figure 1 — Figure 1: The two pipelines. The classifier tags a whole line in one shortest-path search (Eq. 1 ); the tagged form of the TN example is shown between the rows. In the ITN direction, each class grammar is composed with the flexible-join transducer $J$ before it enters the classifier.

Panel of Figure 2 — Figure 2: Two paths through the TN classifier lattice for “ \khmerfont តម្លៃ $5”. Numbers under the states are character offsets. Plain text costs 1 per character and a semiotic token a flat weight, so the path that covers more text with one token wins. No path can leave the digit 5 as plain text: digits are not allowed in name tokens.

$w_{c}$	TN classes	ITN classes
0.30	whitelist
0.40	date, time, telephone,	date, time, telephone,
	electronic, year range,	electronic
	formation
0.45	money, measure, ordinal,	money, measure,
	license plate, roman	license plate

Table 2: Class weights

w_{c}

. A semiotic token costs

w_{c}

regardless of its length; plain text costs 1 per character.

Last words	Repeats	Example
any	$w_{3}$	\khmerfont ផ្ទះធំៗ → \khmerfont ផ្ទះធំ ធំ
$w_{2}$ is \khmerfont មួយ, \khmerfont ទាំង or \khmerfont លើក	$w_{2}w_{3}$	\khmerfont មួយចានៗ → \khmerfont មួយចាន មួយចាន
$w_{3}$ is \khmerfont ម្ដង, \khmerfont ឡើង, \khmerfont ទៅ or \khmerfont ទៀត	$w_{2}w_{3}$	\khmerfont បន្តិចម្ដងៗ → \khmerfont បន្តិចម្ដង បន្តិចម្ដង
and $w_{2}$ a unit of time	$w_{1}w_{2}w_{3}$	\khmerfont ពីរថ្ងៃម្ដងៗ → \khmerfont ពីរថ្ងៃម្ដង ពីរថ្ងៃម្ដង
and $w_{2}$ is \khmerfont ពេល	$w_{3}$	\khmerfont ពេលទៅៗ → \khmerfont ពេលទៅ ទៅ
\khmerfont ម្នាក់ and \khmerfont ម្ដង, either order	$w_{2}w_{3}$	\khmerfont ម្នាក់ម្ដងៗ → \khmerfont ម្នាក់ម្ដង ម្នាក់ម្ដង

Table 3: How much text \khmerfont ៗ repeats, decided on the last three words

w_{1}w_{2}w_{3}

before it as segmented by khmercut. Units of time are \khmerfont ម៉ោង, \khmerfont ថ្ងៃ, \khmerfont យប់, \khmerfont សប្ដាហ៍, \khmerfont អាទិត្យ, \khmerfont ខែ and \khmerfont ឆ្នាំ.

Panel of Figure 3 — Figure 3: The ITN cardinal grammar writes every number as 15 zero-padded digits, one block per scale word, then removes the leading zeros. Each block accepts the standard and the colloquial spoken structures.

Figure 4: Size of each class grammar before it is combined into the classifier. Money and measure are the largest because every currency and unit is its own path. Classes without a bar in one direction exist only in the other.

		$n$	before	after
TN	cardinal (exact)	274	236
	cardinal ( \khmerfont ពិល $\equiv$ \khmerfont ពីរ)	274	274
ITN	cardinal	265	228	265
	decimal	27	19	23
	time	13	7	13

Table 4: Agreement with the test suite of Google’s Khmer grammars [ 24 ] . ITN columns: correct before and after adding the spelling \khmerfont ប្រាំពិល (7) to the ITN grammar.

Figure 5: Semiotic tokens written by ITN in the 2,906 TTS prompts.

Spoken input (excerpt)	Tha output
\khmerfont មួយម៉ែត្រការេ ( \khmerfont ការ៉េ misspelled)	1 m \khmerfont ការេ
\khmerfont …គីឡូម៉ែតក្រឡា ( \khmerfont ម៉ែត្រ misspelled)	574 kg \khmerfont ម៉ែត…
\khmerfont ប្រាំហិកតារ ( \khmerfont ហិកតា misspelled)	5 ha \khmerfont រ
\khmerfont ពីរពាន់ដប់ប្រាំវ៉ុលវ៉ូ (Volvo)	2015 V \khmerfont វ៉ូ
\khmerfont ទម្ងន់មួយរយផោន (weight)	\khmerfont ទម្ងន់£100

Table 5: All ITN errors in the TTS prompts.

Figure 6: Round trip accuracy (%) by class and by how the spoken words are joined: glued, with (U+2581), or with zero-width spaces (U+200B). 300 tokens per class.

Figure 7: Ablation of the ITN design choices. Left: lexicon entries that ITN changes, by where the number’s span falls relative to khmercut’s word boundaries; “inside a word” is an error. Right: TTS prompt sentences changed.

Figure 8: Time per sentence against sentence length. Lines are medians over 20-character bins.

Low-Resource Language Processing

Low-resource language processing builds and evaluates language technologies for languages with limited data, tools, or digital representation.

See all 565 papers

VietNormalizer: An Open-Source, Dependency-Free Python Library for Vietnamese Text Normalization in TTS and NLP Applications
Mar 4, 2026Hung Vu Nguyen, Loan Do, Thanh Ngoc Nguyen +5Low-Resource Language Processing
Australian Catholic University (ACU) · FPT University · ICMS +4
BanglaKontho: Closing the Long-Form Gap in Bangla Text-to-Speech
Sep 24, 2026Mizbaul Haque MarufTTS Synthesis Low-Resource TTS Synthesis
Vivasoft Limited, Dhaka, Bangladesh
Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech
Sep 3, 2026Kunat Pipatanakul, Potsawee Manakul, Warit Sirichotedumrong +3TTS Synthesis TTS Evaluation
Wayu Research · Paxa Labs · Typhoon

Mar 4, 2026J/K move · Enter open · S save

VietNormalizer: An Open-Source, Dependency-Free Python Library for Vietnamese Text Normalization in TTS and NLP Applications

Hung Vu Nguyen, Loan Do, Thanh Ngoc Nguyen, Ushik Shrestha Khwakhali+4

Australian Catholic University (ACU) · FPT University · ICMS+4

Low-Resource Language Processing

Open paper AskPDF arXiv

We present VietNormalizer1, an open-source, zero-dependency Python library for Vietnamese text normalization targeting Text-to-Speech (TTS) and Natural Language Processing (NLP) applications. Vietnamese text normalization is a critical yet underserved preprocessing step: real-world Vietnamese text is densely populated with non-standard words (NSWs), including numbers, dates, times, currency amounts, percentages, acronyms, and foreign-language terms, all of which must be converted to fully pronounceable Vietnamese words before TTS synthesis or downstream language processing. Existing Vietnamese normalization tools either require heavy neural dependencies while covering only a narrow subset of NSW classes, or are embedded within larger NLP toolkits without standalone installability. VietNormalizer addresses these gaps through a unified, rule-based pipeline that: (1) converts arbitrary integers, decimals, and large numbers to Vietnamese words; (2) normalizes dates and times to their spoken Vietnamese forms; (3) handles VND and USD currency amounts; (4) expands percentages; (5) resolves acronyms via a customizable CSV dictionary; (6) transliterates non-Vietnamese loanwords and foreign terms to Vietnamese phonetic approximations; and (7) performs Unicode normalization and emoji/special-character removal. All regular expression patterns are pre-compiled at initialization, enabling high-throughput batch processing with minimal memory overhead and no GPU or external API dependency. The library is installable via pip install vietnormalizer, available on PyPI and GitHub at https://github.com/nghimestudio/vietnormalizer, and released under the MIT license. We discuss the design decisions, limitations of existing approaches, and the generalizability of the rule-based normalization paradigm to other low-resource tonal and agglutinative languages.

Abstract

Low-Resource Language Processing

Explore similar work

VietNormalizer: An Open-Source, Dependency-Free Python Library for Vietnamese Text Normalization in TTS and NLP Applications

BanglaKontho: Closing the Long-Form Gap in Bangla Text-to-Speech

Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech