feat(train): Mixer-TTS training (Lightning) by JarbasAl · Pull Request #146 · TigreGotico/phoonnx

JarbasAl · 2026-06-05T18:37:19Z

Adds Mixer-TTS training to phoonnx_train, ported from nipponjo/mixer-tts-pytorch into phoonnx_train's Lightning framework (complements the inference engine in the merged Mixer-TTS PR).

What's in

phoonnx_train/mixertts/ — vendored pure-torch model + losses + unsupervised aligner + dataset (rewritten to the package namespace).
MixerTTSModel(LightningModule) — wraps the model's training forward + _metrics (mel/duration/pitch/energy/CTC/binarization losses) and an optional LSGAN PatchDiscriminator for mel naturalness. Manual optimization with generator + discriminator optimizers, matching the upstream GAN loop.
train CLI — python -m phoonnx_train.mixertts.train --dataset-dir … --quality {x-low,medium,high} (1.74M/3.17M/20.6M params).
[train] extra gains einops + numba (the aligner).
Docs: training section in docs/mixertts.md.

Verified

4 smoke tests — the Lightning module builds the model + GAN critic, exposes two optimizers (one without GAN), and the inference/export path runs. Full suite 215 passed.

Scope: this wires up the training pipeline + CLI. Full convergence needs a preprocessed dataset (mel + pitch dirs) and a GPU — out of scope for CI. A trained checkpoint exports to ONNX via the same contract as the mirrored voices.

🤖 Generated with Claude Code

Add phoonnx_train/mixertts/ — Mixer-TTS training ported from nipponjo/mixer-tts-pytorch into phoonnx_train's Lightning framework: - MixerTTSModel(LightningModule): wraps the model's training forward + _metrics (mel/duration/pitch/energy/CTC/binarization losses + unsupervised aligner) and an optional LSGAN PatchDiscriminator (manual optimization, generator + disc optimizers), matching the upstream GAN loop. - vendored pure-torch model + losses + aligner + dataset under mixertts/models and mixertts/utils (rewritten to the package namespace). - train CLI (phoonnx_train.mixertts.train) with quality tiers (1.74M/3.17M/20.6M). - [train] extra gains einops + numba (aligner). 4 smoke tests (builds, g+d optimizers, no-gan single optimizer, inference path); full suite 215 passed. Convergence needs a preprocessed dataset + GPU. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

coderabbitai · 2026-06-05T18:37:26Z

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 09925b58-0e13-47cb-bf9c-f5a5d4463479

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch feat/mixertts-training

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

github-actions · 2026-06-05T18:37:55Z

Look what I found! The automated check results are in. 🔍

I've aggregated the results of the automated checks for this PR below.

📋 Repo Health

Scanning for any signs of 'large file' weight gain. ⚖️

⚠️ Some required files are missing.

Latest Version: 1.10.0a1

✅ phoonnx/version.py — Version file
✅ README.md — README
❌ LICENSE — License file
✅ pyproject.toml — pyproject.toml
⚠️ setup.py — setup.py
✅ CHANGELOG.md — Changelog
✅ phoonnx/version.py has valid version block markers

🏷️ Release Preview

I've generated a preview of the upcoming changes. 🎬

Current: 1.10.0a1 → Next: 1.10.0a2

Signal	Value
Label	(none)
PR title	`feat(train): Mixer-TTS training (Lightning)`
Bump	alpha

⚠️ No conventional commit prefix — alpha-only bump.
Suggested: fix: update the thing or feat: update the thing

🚀 Release Channel Compatibility

Predicted next version: 1.10.0a2

Channel	Status	Note	Current Constraint
Stable	⚪	Not in channel	-
Testing	⚪	Not in channel	-
Alpha	⚪	Not in channel	-

📊 Coverage

Diving deep into the code to see what's covered! 🤿

❌ 39.4% total coverage

Files below 80% coverage (37 files)

File	Coverage	Missing lines
`phoonnx/cli.py`	0.0%	98
`phoonnx/thirdparty/kog2p/__init__.py`	0.0%	203
`phoonnx/thirdparty/mantoq/unicode_symbol2label.py`	0.0%	1
`phoonnx/thirdparty/bw2ipa.py`	7.5%	86
`phoonnx/thirdparty/mantoq/pyarabic/number.py`	7.7%	371
`phoonnx/thirdparty/mantoq/buck/phonetise_buckwalter.py`	10.4%	180
`phoonnx/thirdparty/hangul2ipa.py`	16.6%	372
`phoonnx/phonemizers/en.py`	17.5%	104
`phoonnx/thirdparty/mantoq/pyarabic/trans.py`	18.2%	135
`phoonnx/model_manager.py`	20.0%	212
`phoonnx/voice.py`	21.7%	220
`phoonnx/thirdparty/zh_num.py`	23.1%	83
`phoonnx/phonemizers/mul.py`	23.9%	236
`phoonnx/thirdparty/tashkeel/__init__.py`	23.9%	89
`phoonnx/phonemizers/zh.py`	27.0%	92
`phoonnx/phonemizers/ko.py`	30.4%	32
`phoonnx/phonemizers/gl.py`	31.1%	42
`phoonnx/phonemizers/ar.py`	31.2%	44
`phoonnx/thirdparty/mantoq/buck/tokenization.py`	32.5%	27
`phoonnx/thirdparty/phonikud/__init__.py`	35.3%	11
`phoonnx/phonemizers/ja.py`	36.0%	32
`phoonnx/phonemizers/fa.py`	36.4%	14
`phoonnx/phonemizers/pt.py`	38.1%	13
`phoonnx/thirdparty/mantoq/pyarabic/normalize.py`	38.1%	13
`phoonnx/thirdparty/mantoq/pyarabic/araby.py`	39.7%	298
`phoonnx/phonemizers/he.py`	40.0%	12
`phoonnx/phonemizers/vi.py`	40.0%	12
`phoonnx/phonemizers/base.py`	40.8%	71
`phoonnx/thirdparty/mantoq/pyarabic/stack.py`	45.5%	6
`phoonnx/thirdparty/mantoq/num2words.py`	47.6%	11
`phoonnx/phonemizers/mwl.py`	50.0%	8
`phoonnx/tokenizer.py`	52.4%	147
`phoonnx/thirdparty/mantoq/__init__.py`	60.0%	10
`phoonnx/thirdparty/mantoq/pyarabic/arabrepr.py`	60.0%	6
`phoonnx/config.py`	61.0%	130
`phoonnx/engines/vocoders/griffinlim.py`	61.4%	27
`phoonnx/engines/optispeech.py`	69.6%	24

Full report: download the coverage-report artifact.

🔍 Lint

Ensuring your contribution is moving forward. 🚀

❌ ruff: issues found — see job log

🔒 Security (pip-audit)

Ensuring our encryption is top-notch. 🔐

✅ No known vulnerabilities found (61 packages scanned).

⚖️ License Check

Checking for any restrictive patent clauses. 📜

❌ License violations detected (43 packages) — review required before merging.

Dependency                          License Name                                            License Type         Misc                                    
phoonnx:1.3.3                       Error                                                   Error                                                        

License Type                        Found                                                  
Error                               1

License distribution: 14× MIT License, 7× Apache Software License, 5× MIT, 3× Apache-2.0, 2× BSD-3-Clause, 2× ISC License (ISCL), 1× 3-Clause BSD License, 1× Apache Software License; BSD License, +8 more

Full breakdown — 43 packages

Package	Version	License	URL
`build`	1.5.0	MIT	link
`certifi`	2026.5.20	Mozilla Public License 2.0 (MPL 2.0)	link
`charset-normalizer`	3.4.7	MIT	link
`click`	8.4.1	BSD-3-Clause	link
`combo_lock`	0.3.1	Apache-2.0	link
`dateparser`	1.4.0	BSD License	link
`filelock`	3.29.1	MIT	link
`flatbuffers`	25.12.19	Apache Software License	link
`idna`	3.18	BSD-3-Clause	link
`json-database`	0.10.1	MIT	link
`kthread`	0.2.3	MIT License	link
`langcodes`	3.5.1	MIT License	link
`markdown-it-py`	4.2.0	MIT License	link
`mdurl`	0.1.2	MIT License	link
`memory-tempfile`	2.2.3	MIT License	link
`numpy`	2.4.6	BSD-3-Clause AND 0BSD AND MIT AND Zlib AND CC0-1.0	link
`onnxruntime`	1.26.0	MIT License	link
`ovos-config`	2.1.1	Apache-2.0	link
`ovos-date-parser`	0.7.0a5	Apache Software License	link
`ovos-number-parser`	0.5.1	Apache Software License	link
`ovos-utils`	0.8.5	Apache-2.0	link
`packaging`	26.2	Apache-2.0 OR BSD-2-Clause	link
`pexpect`	4.9.0	ISC License (ISCL)	link
`phoonnx`	1.10.0a1	Apache Software License	link
`protobuf`	7.35.0	3-Clause BSD License	link
`ptyprocess`	0.7.0	ISC License (ISCL)	link
`pyee`	13.0.1	MIT License	link
`Pygments`	2.20.0	BSD-2-Clause	link
`pyproject_hooks`	1.2.0	MIT License	link
`python-dateutil`	2.9.0.post0	Apache Software License; BSD License	link
`pytz`	2026.2	MIT License	link
`PyYAML`	6.0.3	MIT License	link
`quebra-frases`	0.3.7	Apache Software License	link
`regex`	2026.5.9	Apache-2.0 AND CNRI-Python	link
`requests`	2.34.2	Apache Software License	link
`rich`	13.9.4	MIT License	link
`rich-click`	1.9.8	MIT License

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.

Policy: Apache 2.0 (universal donor). StrongCopyleft / NetworkCopyleft / WeakCopyleft / Other / Error categories fail. MPL allowed.

🔨 Build Tests

Checking the plumbing of your data flows. 🚰

✅ All versions pass

Python	Build	Install	Tests
3.10	✅	✅	✅
3.11	✅	✅	✅
3.12	✅	✅	✅
3.13	✅	✅	✅
3.14	✅	✅	✅

From the digital workshop of OpenVoiceOS. 🛠️

JarbasAl · 2026-06-08T13:23:28Z

Companion tracking issue: #169

github-actions Bot added the feature label Jun 5, 2026

This was referenced Jun 8, 2026

MixerTTS inference engine #161

Closed

Mixer-TTS training engine #169

Open

Pluggable multi-engine training framework #177

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(train): Mixer-TTS training (Lightning)#146

feat(train): Mixer-TTS training (Lightning)#146
JarbasAl wants to merge 1 commit into
devfrom
feat/mixertts-training

JarbasAl commented Jun 5, 2026

Uh oh!

coderabbitai Bot commented Jun 5, 2026

Review skipped

Uh oh!

github-actions Bot commented Jun 5, 2026 •

edited

Loading

Uh oh!

JarbasAl commented Jun 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

JarbasAl commented Jun 5, 2026

What's in

Verified

Uh oh!

coderabbitai Bot commented Jun 5, 2026

Review skipped

Uh oh!

github-actions Bot commented Jun 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Look what I found! The automated check results are in. 🔍

📋 Repo Health

🏷️ Release Preview

📊 Coverage

🔍 Lint

🔒 Security (pip-audit)

⚖️ License Check

🔨 Build Tests

Uh oh!

JarbasAl commented Jun 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

github-actions Bot commented Jun 5, 2026 •

edited

Loading