techiaith-speech-normalizer 0.2.0__tar.gz

techiaith_speech_normalizer-0.2.0/.github/build-requirements.in

@@ -0,0 +1,4 @@
+# Tools that .github/workflows/publish.yml uses to build the sdist and wheel.
+# build-requirements.txt pins them with hashes; Dependabot keeps it up to date.
+build
+hatchling

techiaith_speech_normalizer-0.2.0/.github/build-requirements.txt

@@ -0,0 +1,40 @@
+#
+# This file is autogenerated by pip-compile with Python 3.12
+# by the following command:
+#
+#    pip-compile --generate-hashes --output-file=.github/build-requirements.txt --strip-extras .github/build-requirements.in
+#
+build==1.6.1 \
+    --hash=sha256:51cc11666391ab6f092070437ac747002ff46f3e4113a3622177ee6b488bfc53 \
+    --hash=sha256:ecd351a4be9d35a9eaaba244a7687143c9c7d4aea6ac964e7e7ddab20cbcf4e7
+    # via -r .github/build-requirements.in
+hatchling==1.33.0 \
+    --hash=sha256:415290b84f464d2884303fca8974d38e7b85579069fab02a15db0828fbb1446d \
+    --hash=sha256:4c422bd4d554fcff5de5b73e9596f85e41e9f97f7e553d8f7595ecb31a627bef
+    # via -r .github/build-requirements.in
+packaging==26.3 \
+    --hash=sha256:94edc256424af38762eb31306eed28beb9f0efc50a8837492c9d6fd6004aed79 \
+    --hash=sha256:d7193f7c8e4e93f444fde0262bf90af30e16fa0ad0ad44cb553c87339b23cd1c
+    # via
+    #   build
+    #   hatchling
+pathspec==1.1.1 \
+    --hash=sha256:17db5ecd524104a120e173814c90367a96a98d07c45b2e10c2f3919fff91bf5a \
+    --hash=sha256:a00ce642f577bf7f473932318056212bc4f8bfdf53128c78bbd5af0b9b20b189
+    # via hatchling
+pluggy==1.7.0 \
+    --hash=sha256:7dd7b0d8832ba3cb632c306926ded123429211b83641b35dc5c41ad2d34f9bec \
+    --hash=sha256:d1eaa46ebb595891b860ab086b4d09c8588af65ebd4361b8e8f4bb8920b90ba8
+    # via hatchling
+pyproject-hooks==1.3.3 \
+    --hash=sha256:5fc53fdac9f7bd63fbcdc868fb5f90b4784d78a53a3d3388cd738b807441a20b \
+    --hash=sha256:defda19b854fa0d3bd4f76ea4ddcba8abd7dcfcdd585a6690ade050744fc5f43
+    # via build
+tomlkit==0.15.2 \
+    --hash=sha256:05562412163a3037b94497385fb7f9204c637bc605af7a48199e41cf2272f437 \
+    --hash=sha256:83e32755ba1d20d9340bd069ee97e9f6232ba2346b63ffcba425de01f7e327a6
+    # via hatchling
+trove-classifiers==2026.9.21.13 \
+    --hash=sha256:0a9ebc8d4e2f3e8a22848c5258033035bec17a3012ac3fea16dbaa764489eb71 \
+    --hash=sha256:8b1ff4f9c191b1040b71c37f1e445ab99732911e3cd91de52838453a854d7a17
+    # via hatchling

techiaith_speech_normalizer-0.2.0/.github/dependabot.yml

@@ -0,0 +1,23 @@
+# The cooldown holds back new releases for a week, so a compromised release is
+# likely to be found and pulled before it reaches this repository.
+version: 2
+updates:
+  - package-ecosystem: github-actions
+    directory: /
+    schedule:
+      interval: weekly
+    cooldown:
+      default-days: 7
+    groups:
+      github-actions:
+        patterns: ["*"]
+
+  - package-ecosystem: pip
+    directory: /.github
+    schedule:
+      interval: weekly
+    cooldown:
+      default-days: 7
+    groups:
+      build-tools:
+        patterns: ["*"]

techiaith_speech_normalizer-0.2.0/.github/workflows/ci.yml

@@ -0,0 +1,61 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    name: Test (Python ${{ matrix.python-version }})
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.9", "3.10", "3.11", "3.12", "3.13", "3.14"]
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+      - uses: actions/setup-python@6d4fd22b20465ccc0f7354476588e8df7c9ca0e0 # v7.1.0
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Install
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install -e . pytest
+      - name: Test
+        run: python -m pytest -q
+
+  build:
+    name: Build distribution
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+      - uses: actions/setup-python@6d4fd22b20465ccc0f7354476588e8df7c9ca0e0 # v7.1.0
+        with:
+          python-version: "3.12"
+      # Builds the same way as publish.yml, then checks the metadata.
+      - name: Build sdist and wheel
+        run: |
+          python -m pip install --require-hashes --only-binary :all: -r .github/build-requirements.txt
+          python -m build --no-isolation
+          python -m pip install twine
+          python -m twine check --strict dist/*
+
+  zizmor:
+    name: Check workflows (zizmor)
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+      - uses: zizmorcore/zizmor-action@53290badfa369cb0bf7ad16342cb62c2fa47ac7b # v0.8.0
+        with:
+          advanced-security: false

techiaith_speech_normalizer-0.2.0/.github/workflows/publish.yml

@@ -0,0 +1,120 @@
+# Publishes a release to PyPI and GitHub when a version tag (e.g. v0.3.0) is
+# pushed. Each job gets only the permissions it needs:
+#
+#   build           checks the tag and builds the sdist and wheel with the
+#                   hash-pinned tools in .github/build-requirements.txt;
+#                   read-only token, no secrets
+#   publish-pypi    uploads to PyPI with trusted publishing (OpenID Connect), so
+#                   no API token is stored anywhere; runs in the "pypi"
+#                   environment, which waits for a maintainer's approval
+#   github-release  attests the files' build provenance and creates the GitHub
+#                   release with the files and the notes from CHANGELOG.md
+#
+# PyPI trusted publisher: owner "techiaith", repository
+# "techiaith-speech-normalizer", workflow "publish.yml", environment "pypi".
+# See https://docs.pypi.org/trusted-publishers/
+
+name: Publish
+
+on:
+  push:
+    tags: ["v*"]
+
+permissions: {}
+
+jobs:
+  build:
+    name: Build distribution
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          fetch-depth: 0 # to check that the tagged commit is on main
+          persist-credentials: false
+      - uses: actions/setup-python@6d4fd22b20465ccc0f7354476588e8df7c9ca0e0 # v7.1.0
+        with:
+          python-version: "3.12"
+      - name: Check the tag against main and the package version
+        run: |
+          if ! git merge-base --is-ancestor "$GITHUB_SHA" origin/main; then
+            echo "$GITHUB_REF_NAME does not point to a commit on main" >&2
+            exit 1
+          fi
+          version=$(python -c 'import tomllib; print(tomllib.load(open("pyproject.toml", "rb"))["project"]["version"])')
+          if [ "$version" != "${GITHUB_REF_NAME#v}" ]; then
+            echo "pyproject.toml version ($version) does not match the tag ($GITHUB_REF_NAME)" >&2
+            exit 1
+          fi
+          citation=$(sed -n 's/^version: *//p' CITATION.cff | tr -d "\"'")
+          if [ "$citation" != "$version" ]; then
+            echo "CITATION.cff version ($citation) does not match pyproject.toml ($version)" >&2
+            exit 1
+          fi
+          if ! grep -q "^## \[$version\]" CHANGELOG.md; then
+            echo "CHANGELOG.md has no section for $version" >&2
+            exit 1
+          fi
+      - name: Build sdist and wheel
+        run: |
+          python -m pip install --require-hashes --only-binary :all: -r .github/build-requirements.txt
+          python -m build --no-isolation
+      - uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
+        with:
+          name: dist
+          path: dist/
+          if-no-files-found: error
+
+  publish-pypi:
+    name: Publish to PyPI
+    needs: build
+    runs-on: ubuntu-latest
+    environment:
+      name: pypi
+      url: https://pypi.org/p/techiaith-speech-normalizer
+    permissions:
+      id-token: write # trusted publishing and PEP 740 attestations
+    steps:
+      - uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8.0.2
+        with:
+          name: dist
+          path: dist/
+      - uses: pypa/gh-action-pypi-publish@7245906fb61fa33a13554f9d5fb856fe61758086 # v1.14.3
+
+  github-release:
+    name: Create GitHub release
+    needs: publish-pypi
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write # create the release
+      id-token: write # sign the provenance attestation
+      attestations: write # store the provenance attestation
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+          sparse-checkout: CHANGELOG.md
+          sparse-checkout-cone-mode: false
+      - uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8.0.2
+        with:
+          name: dist
+          path: dist/
+      - uses: actions/attest-build-provenance@4d101475d8b20a2381f78447822ac1eab6504dd8 # v4.2.2
+        with:
+          subject-path: dist/*
+      - name: Create the release
+        env:
+          GH_TOKEN: ${{ github.token }}
+          GH_REPO: ${{ github.repository }}
+        run: |
+          # The notes are this version's section of CHANGELOG.md.
+          awk -v heading="## [${GITHUB_REF_NAME#v}]" '
+            /^## \[/ { if (found) exit; found = (index($0, heading) == 1); next }
+            found
+          ' CHANGELOG.md > "$RUNNER_TEMP/notes.md"
+          # Attach the files while the release is a draft: once published, an
+          # immutable release cannot be changed.
+          gh release create "$GITHUB_REF_NAME" dist/* --draft --verify-tag \
+            --title "$GITHUB_REF_NAME" --notes-file "$RUNNER_TEMP/notes.md"
+          gh release edit "$GITHUB_REF_NAME" --draft=false

techiaith_speech_normalizer-0.2.0/.gitignore

@@ -0,0 +1,14 @@
+__pycache__/
+*.py[cod]
+*.egg-info/
+dist/
+build/
+.eggs/
+*.egg
+.venv/
+venv/
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+.DS_Store
+uv.lock

techiaith_speech_normalizer-0.2.0/CHANGELOG.md

@@ -0,0 +1,47 @@
+# Changelog
+
+All notable changes to this project are documented in this file.
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/)
+and the project follows [Semantic Versioning](https://semver.org/).
+
+## [Unreleased]
+
+## [0.2.0] - 2026-10-29
+
+First public release, on GitHub and PyPI. Changes since 0.1.0:
+
+### Added
+- `WelshNumberNormalizer`: converts Welsh number words to digits (decimal and
+  vigesimal systems, ordinals, currency, mutated forms). Enabled in
+  `WelshNormalizer` with `normalize_numbers=True`.
+- WER-only toggles, all off by default: `lowercase`, `strip_punctuation`,
+  `join_hyphenated_compounds`.
+- Unicode NFKC normalisation as the first step of the pipeline, and a callable
+  `__call__` interface matching OpenAI Whisper's normalizers.
+- Expansion of `{llafar|ysgrifenedig}` annotations, e.g. `{saith|7}` → `7`.
+- `standardize_adjectives` and the `ADJECTIVE_REDUCTIONS` table, plus many new
+  vocabulary entries including possessive forms.
+- Contractions `timod`/`t'mod`/`t'bo'`, `fan'na`, `'na beth yw`, `fel na` and
+  `ma' na`; fixes for `dwi'n`/`dwi'm`, `ma' 'na` → `mae yna`, and tokens with a
+  leading apostrophe (`'di`, `'na`).
+- `COMPARISON.md`, comparing this normaliser with OpenAI Whisper's text
+  normalisers.
+- GitHub Actions workflows for tests and for publishing releases to PyPI and
+  GitHub, plus `CONTRIBUTING.md`, `SECURITY.md` and `CITATION.cff`.
+
+### Changed
+- Package renamed to the `techiaith.speech_normalizer` namespace.
+- Python 3.9 or later is now required; support for Python 3.8 has been
+  dropped.
+- The README now states that `remove_tags` and `remove_asterisks` are off by
+  default (as the code has always behaved), and its examples reflect the
+  current hesitation standardisation (`yy`/`yrm`/`ymm` → `yym`).
+- The tests for tag and asterisk removal now enable those options explicitly,
+  and new tests check that both are kept by default.
+
+## [0.1.0] - 2026-05-19
+
+### Added
+- Initial release: configurable `WelshNormalizer` for Welsh verbatim speech
+  transcriptions, SRT subtitle file support, uncertainty flagging and the
+  `techiaith-speech-normalize` command-line tool.

techiaith_speech_normalizer-0.2.0/CITATION.cff

@@ -0,0 +1,22 @@
+cff-version: 1.2.0
+message: "If you use this software, please cite it as below."
+title: "techiaith-speech-normalizer: Normaleiddiwr Lleferydd Cymraeg / Welsh Speech Normalizer"
+type: software
+authors:
+  - family-names: Jones
+    given-names: Dewi Bryn
+    affiliation: "Uned Technolegau Iaith, Prifysgol Bangor / Language Technologies Unit, Bangor University"
+  - family-names: Wanasky
+    given-names: Sasha
+    affiliation: "Prifysgol Bangor / Bangor University"
+repository-code: "https://github.com/techiaith/techiaith-speech-normalizer"
+license: MIT
+version: 0.2.0
+date-released: "2026-10-29"
+keywords:
+  - welsh
+  - cymraeg
+  - speech recognition
+  - transcription
+  - text normalization
+  - nlp

techiaith_speech_normalizer-0.2.0/COMPARISON.md

@@ -0,0 +1,162 @@
+# WelshNormalizer vs. Whisper Text Normalizers: Comparison & Future Work
+
+## Overview
+
+This document compares `techiaith.speech_normalizer.WelshNormalizer` with OpenAI's Whisper text normalization classes: `BasicTextNormalizer` and `EnglishTextNormalizer`. The goal is to assess whether WelshNormalizer occupies the same role for Welsh that EnglishTextNormalizer does for English.
+
+## Whisper's Normalizers
+
+### BasicTextNormalizer (language-agnostic)
+
+A minimal, language-agnostic normalizer that prepares text for WER (Word Error Rate) comparison. Its pipeline:
+
+1. **Lowercasing**
+2. **Annotation removal** — strips text within `<...>`, `[...]`, and `(...)`
+3. **Unicode normalization** — NFKD (with diacritic removal) or NFKC (preserving diacritics)
+4. **Symbol/punctuation stripping** — replaces all Unicode Mark, Symbol, and Punctuation characters with spaces
+5. **Optional grapheme splitting** — splits into individual grapheme clusters
+6. **Whitespace collapsing**
+
+This normalizer performs no linguistic transformations — no contraction expansion, no spelling correction, no number handling. It is purely a text-cleaning pass.
+
+### EnglishTextNormalizer (English-specific)
+
+A linguistically-aware normalizer for English speech transcription. Its pipeline:
+
+1. **Lowercasing**
+2. **Annotation removal** — same as BasicTextNormalizer
+3. **Filler word removal** — removes `hmm`, `mm`, `mhm`, `mmm`, `uh`, `um`
+4. **Contraction expansion** — 40+ patterns (`won't` → `will not`, `gonna` → `going to`, `'re` → `are`, etc.)
+5. **Abbreviation expansion** — titles and honorifics (`mr` → `mister`, `dr` → `doctor`, `prof` → `professor`, etc.)
+6. **Number-to-digit conversion** — a complex stateful sliding-window parser handling:
+   - Cardinal and ordinal numbers (`twenty million` → `20000000`, `third` → `3rd`)
+   - Currency (`twenty dollars` → `$20`, combining dollars and cents)
+   - Decimals (`three point five` → `3.5`)
+   - Nominal sequences (`double oh seven` → `007`)
+   - Percentage (`fifty percent` → `50%`)
+7. **British → American spelling** — dictionary lookup from a JSON file
+8. **Symbol cleanup and whitespace collapsing**
+
+## Feature-by-Feature Comparison
+
+| Capability | BasicText | EnglishText | WelshNormalizer |
+|---|:---:|:---:|:---:|
+| Lowercasing | Yes | Yes | Optional toggle |
+| Remove annotations/tags | `<>` `[]` `()` | `<>` `[]` `()` | `<>` tags |
+| Unicode normalization (NFKD/NFKC) | Yes | Yes | Yes (NFKC) |
+| Strip all punctuation/symbols | Yes | Selective | Optional toggle |
+| Hesitation/filler removal | No | Yes | Yes |
+| Hesitation standardization | No | No | Yes |
+| Contraction expansion | No | ~40 patterns | ~130+ patterns |
+| Pronoun normalization | No | No | Yes |
+| Dialect → standard spelling | No | British → American | Dialect → Standard Welsh |
+| Number-to-digit parsing | No | Yes (complex) | Yes (decimal, vigesimal, mutated forms) |
+| Currency normalization | No | Yes | Yes (punt/ceiniog) |
+| Abbreviation expansion (titles) | No | Yes | No |
+| Multi-word phrase correction | No | No | Yes |
+| Context-aware transforms | No | Partial | Yes |
+| Uncertainty flagging | No | No | Yes |
+| Per-category toggles | Minimal | No | Yes (22 toggles) |
+| SRT subtitle file support | No | No | Yes |
+| Custom user corrections | No | No | Yes |
+| False start removal | No | No | Yes |
+| Hyphenated compound joining | No | No | Optional toggle |
+
+## Where WelshNormalizer Matches or Exceeds
+
+### Contraction and dialect handling
+
+EnglishTextNormalizer handles approximately 40 contraction patterns. WelshNormalizer handles 130+ across multiple categories: pronouns, verb contractions, `mae` contractions, other contractions, dialectal plurals (North `-a` and South `-e` → standard `-au`), dialectal verbs, internal vowel reductions, and spelling corrections. This reflects the greater distance between spoken and written Welsh compared to English.
+
+### Context-sensitive rules
+
+Whisper's English normalizer uses ordered regex patterns to disambiguate contractions (e.g., matching `'d been` → `had been` before the generic `'d` → `would`). WelshNormalizer implements explicit contextual logic:
+
+- **`'di` → `wedi`** only when preceded by a pronoun (`dwi 'di` → `dwi wedi`), left unchanged otherwise
+- **`ma'` + follower** normalization using a lookup table that normalizes both `ma'` and its following word together
+
+These are genuinely context-sensitive transformations that go beyond pattern ordering.
+
+### Uncertainty flagging
+
+WelshNormalizer flags ambiguous cases for human review rather than guessing. For example, `'di` without a preceding pronoun is flagged as `CONTEXT_UNCERTAIN`. Whisper's normalizers have no equivalent — they always produce a deterministic output with no indication of confidence.
+
+### Configurability
+
+Whisper's normalizers are essentially all-or-nothing (`BasicTextNormalizer` has two boolean options; `EnglishTextNormalizer` has none). WelshNormalizer offers 18 independent toggles, allowing researchers to measure the impact of each normalization category on downstream tasks like WER calculation.
+
+### Additional features
+
+- **SRT file processing** — direct support for normalizing subtitle files
+- **Custom corrections** — users can supply their own single-word and multi-word replacements
+- **False start removal** — removes truncated words (`m-`, `o-`) common in spontaneous speech
+
+## Where WelshNormalizer Falls Short
+
+### ~~1. Unicode normalization~~ (DONE)
+
+NFKC normalization is now applied as the first step in the pipeline, ensuring circumflexed vowels (`ŵ`, `â`, etc.) match lookup tables regardless of encoding.
+
+### ~~2. Number handling~~ (DONE)
+
+A separate `WelshNumberNormalizer` class now handles:
+
+- **Decimal system**: `tri deg pump` → `35`
+- **Vigesimal system**: `pymtheg ar hugain` → `35`
+- **Teens**: `un ar ddeg` → `11`, `deunaw` → `18`
+- **Multipliers**: `mil naw cant pedwar deg tri` → `1943`
+- **Ordinals**: `trydydd` → `3ydd`, `cyntaf` → `1af`
+- **Currency**: `pum punt a deg ceiniog` → `£5.10`
+- **Feminine forms**: `dwy`, `tair`, `pedair`, `trydedd`, etc.
+- **Mutated forms**: soft mutation (`ddau`, `dri`, `fil`, `gyntaf`, etc.), aspirate mutation (`thri`, `phum`, `phedwar`, etc.)
+
+Integrated into WelshNormalizer via an optional `normalize_numbers` toggle (off by default), following Whisper's compositional sub-normalizer pattern.
+
+### 3. Abbreviation and title expansion
+
+**Gap:** No expansion of titles or common abbreviations.
+
+**Impact:** Transcriptions may contain `Mr`, `Mrs`, `Dr`, `St` (sant/stryd), `Prif`, etc. These are not expanded.
+
+**Recommendation:** Lower priority than number handling. Could be added as an additional lookup table in `welsh_data.py` if needed. Note that some Welsh abbreviations are more complex due to mutation (`Dr` → `Doctor`/`Ddoctor` depending on context).
+
+### ~~4. Lowercasing~~ (DONE)
+
+Optional `lowercase` toggle added (off by default). When enabled, converts output to lowercase for WER calculation.
+
+### ~~5. Punctuation and symbol handling~~ (DONE)
+
+Optional `strip_punctuation` toggle added (off by default). When enabled, strips all punctuation and symbols for WER calculation.
+
+## Architectural Comparison
+
+| Aspect | Whisper normalizers | WelshNormalizer |
+|---|---|---|
+| Callable pattern | `__call__` (normalizer as function) | `__call__` + `.normalize()` |
+| Composition | Composes sub-normalizers as classes | Composed `WelshNumberNormalizer` sub-normalizer |
+| Data storage | Inline dictionaries + JSON file | Separate `welsh_data.py` module |
+| Configuration | Constructor booleans (2-3) | Constructor booleans (18) |
+| Return type | `str` | `str` or `Tuple[str, List[Flag]]` |
+| Tokenization | Regex-based inline | Separate tokenizer module |
+
+### Architectural alignment (DONE)
+
+1. **`__call__`** — `WelshNormalizer` is now callable: `normalizer(text)` works alongside `normalizer.normalize(text)`.
+2. **Sub-normalizer composition** — `WelshNumberNormalizer` is a separate class composed into `WelshNormalizer`, following Whisper's pattern.
+
+## Summary
+
+WelshNormalizer is a **WelshTextNormalizer** in the same sense that Whisper's `EnglishTextNormalizer` is an English-specific speech normalization layer. It occupies the same architectural role, handles the same category of problem (spoken-form → written-form normalization), and in several areas (contraction coverage, context sensitivity, configurability, uncertainty handling) exceeds what Whisper provides for English.
+
+With NFKC Unicode normalization, number-to-digit conversion (including vigesimal and mutated forms), `__call__` API alignment, lowercasing, and punctuation stripping now implemented, **WelshNormalizer has reached effective parity with EnglishTextNormalizer**. The only remaining gap is title/abbreviation expansion, which is a minor completeness improvement.
+
+## Priority Roadmap
+
+| Priority | Task | Effort | Impact | Status |
+|---|---|---|---|---|
+| 1 | Add NFKC Unicode normalization pass | Small | Robustness against encoding variation | Done |
+| 2 | Add `__call__` method | Trivial | API consistency with Whisper | Done |
+| 3 | Welsh number normalizer (decimal, vigesimal, mutations) | Large | Feature parity for number handling | Done |
+| 4 | Optional lowercasing toggle | Small | WER calculation support | Done |
+| 5 | Optional punctuation stripping | Small | WER calculation support | Done |
+| 6 | Title/abbreviation expansion | Small | Minor completeness improvement | Pending |

techiaith_speech_normalizer-0.2.0/CONTRIBUTING.md

@@ -0,0 +1,140 @@
+# Cyfrannu / Contributing
+
+## Cymraeg
+
+Croeso i gyfraniadau: adroddiadau am wallau, geiriau ac ymadroddion newydd ar
+gyfer y tablau yn `welsh_data.py`, a gwelliannau i'r cod.
+
+### Gosod ar gyfer datblygu
+
+```bash
+git clone https://github.com/techiaith/techiaith-speech-normalizer.git
+cd techiaith-speech-normalizer
+pip install -e . pytest
+pytest
+```
+
+Neu gydag [uv](https://docs.astral.sh/uv/): `uv run pytest`.
+
+### Ychwanegu geiriau neu reolau newydd
+
+Mae'r holl dablau ieithyddol yn `src/techiaith/speech_normalizer/welsh_data.py`.
+Ar ôl eu newid:
+
+1. Ychwanegwch brawf yn `tests/` ar gyfer y ffurf newydd.
+2. Rhedwch `pytest` a sicrhewch fod pob prawf yn pasio.
+
+### Ceisiadau tynnu (pull requests)
+
+- Agorwch issue yn gyntaf ar gyfer newidiadau mawr.
+- Cadwch bob cais yn fach ac yn canolbwyntio ar un peth.
+- Ychwanegwch nodyn o dan "Unreleased" yn `CHANGELOG.md`.
+
+### Rhyddhau fersiwn newydd (cynhalwyr)
+
+1. Mewn cais tynnu, diweddarwch `version` yn `pyproject.toml`, diweddarwch
+   `version` a `date-released` yn `CITATION.cff`, a symudwch y nodiadau yn
+   `CHANGELOG.md` o "Unreleased" i bennawd y fersiwn newydd, gyda'r dyddiad.
+2. Ar ôl ei uno, tagiwch y commit newydd ar `main` a gwthiwch y tag:
+
+   ```bash
+   git switch main && git pull
+   git tag -a v0.3.0 -m v0.3.0
+   git push origin v0.3.0
+   ```
+
+3. Mae'r llif gwaith `.github/workflows/publish.yml` yn gwirio bod y tag ar
+   `main` a'i fod yn cyfateb i'r fersiwn yn `pyproject.toml`, `CITATION.cff` a
+   `CHANGELOG.md`, ac yna'n adeiladu'r pecyn. Mae cyhoeddi ar PyPI yn aros am
+   gymeradwyaeth yn yr amgylchedd `pypi` (y botwm "Review deployments" yn
+   rhediad y llif gwaith). Wedyn mae'r llif gwaith yn creu'r datganiad (release)
+   ar GitHub, gyda'r ffeiliau a'r nodiadau o `CHANGELOG.md`.
+
+Dim ond gweinyddwyr yr ystorfa sy'n gallu gwthio tagiau `v*`. Ni ellir newid
+datganiad ar ôl ei gyhoeddi, ar PyPI nac ar GitHub: rhaid rhyddhau fersiwn
+newydd i gywiro camgymeriad.
+
+Mae'r offer sy'n adeiladu'r pecyn wedi'u pinio gyda hashes yn
+`.github/build-requirements.txt`, ac mae Dependabot yn cynnig diweddariadau
+iddynt ac i'r GitHub Actions. I'w newid â llaw, golygwch
+`.github/build-requirements.in` a rhedwch:
+
+```bash
+uvx --python 3.12 --from pip-tools pip-compile --generate-hashes --strip-extras --output-file .github/build-requirements.txt .github/build-requirements.in
+```
+
+Mae cyhoeddi ar PyPI yn defnyddio
+[cyhoeddi dibynadwy (trusted publishing)](https://docs.pypi.org/trusted-publishers/),
+felly nid oes tocyn API yn cael ei gadw yn unman. Mae'r cyhoeddwr dibynadwy ar
+PyPI wedi'i osod i berchennog `techiaith`, ystorfa `techiaith-speech-normalizer`,
+llif gwaith `publish.yml` ac amgylchedd `pypi`; os caiff ffeil y llif gwaith
+neu'r amgylchedd eu hailenwi, bydd cyhoeddi'n methu nes eu diweddaru yno hefyd.
+
+## English
+
+Contributions are welcome: bug reports, new words and phrases for the tables in
+`welsh_data.py`, and improvements to the code.
+
+### Development setup
+
+```bash
+git clone https://github.com/techiaith/techiaith-speech-normalizer.git
+cd techiaith-speech-normalizer
+pip install -e . pytest
+pytest
+```
+
+Or with [uv](https://docs.astral.sh/uv/): `uv run pytest`.
+
+### Adding words or rules
+
+All the linguistic tables live in `src/techiaith/speech_normalizer/welsh_data.py`.
+After changing them:
+
+1. Add a test in `tests/` for the new form.
+2. Run `pytest` and make sure everything passes.
+
+### Pull requests
+
+- Open an issue first for larger changes.
+- Keep each pull request small and focused on one thing.
+- Add a note under "Unreleased" in `CHANGELOG.md`.
+
+### Releasing a new version (maintainers)
+
+1. In a pull request, update `version` in `pyproject.toml`, update `version`
+   and `date-released` in `CITATION.cff`, and move the notes in `CHANGELOG.md`
+   from "Unreleased" to a heading for the new version, with the date.
+2. Once it is merged, tag the new commit on `main` and push the tag:
+
+   ```bash
+   git switch main && git pull
+   git tag -a v0.3.0 -m v0.3.0
+   git push origin v0.3.0
+   ```
+
+3. The `.github/workflows/publish.yml` workflow checks that the tag is on
+   `main` and matches the versions in `pyproject.toml`, `CITATION.cff` and
+   `CHANGELOG.md`, then builds the package. Publishing to PyPI waits for
+   approval in the `pypi` environment (the "Review deployments" button on the
+   workflow run). The workflow then creates the GitHub release, with the files
+   and the notes from `CHANGELOG.md`.
+
+Only repository admins can push `v*` tags. A release cannot be changed once it
+is published, on PyPI or on GitHub: fix a mistake by releasing a new version.
+
+The tools that build the package are pinned with hashes in
+`.github/build-requirements.txt`, and Dependabot proposes updates to them and
+to the GitHub Actions. To change them by hand, edit
+`.github/build-requirements.in` and run:
+
+```bash
+uvx --python 3.12 --from pip-tools pip-compile --generate-hashes --strip-extras --output-file .github/build-requirements.txt .github/build-requirements.in
+```
+
+PyPI publishing uses
+[trusted publishing](https://docs.pypi.org/trusted-publishers/), so no API
+token is stored anywhere. The trusted publisher on PyPI is set to owner
+`techiaith`, repository `techiaith-speech-normalizer`, workflow `publish.yml`
+and environment `pypi`; renaming the workflow file or the environment breaks
+publishing until it is updated there too.

techiaith_speech_normalizer-0.2.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Uned Technolegau Iaith, Prifysgol Bangor / Language Technologies Unit, Bangor University
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.