syscon-tts 0.0.1__tar.gz

syscon_tts-0.0.1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Syscon International, Inc.
+
+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.

syscon_tts-0.0.1/PKG-INFO

@@ -0,0 +1,267 @@
+Metadata-Version: 2.4
+Name: syscon-tts
+Version: 0.0.1
+Summary: Offline neural text-to-speech for the PlantStar APU
+Author: Keith Hostetler
+Author-email: robh@syscon-intl.com
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/SYSCON-International/syscon_tts
+Project-URL: Source, https://github.com/SYSCON-International/syscon_tts
+Project-URL: Issues, https://github.com/SYSCON-International/syscon_tts/issues
+Keywords: tts,text-to-speech,piper,offline,plantstar,apu
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Manufacturing
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Operating System :: MacOS :: MacOS X
+Classifier: Operating System :: Microsoft :: Windows
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Topic :: Multimedia :: Sound/Audio :: Speech
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: piper-tts==1.2.0; sys_platform == "linux" and python_version < "3.12"
+Provides-Extra: piper
+Requires-Dist: piper-tts==1.2.0; extra == "piper"
+Provides-Extra: server
+Requires-Dist: fastapi<1.0,>=0.111; extra == "server"
+Requires-Dist: uvicorn[standard]<1.0,>=0.30; extra == "server"
+Provides-Extra: dev
+Requires-Dist: pytest<9,>=8.2; extra == "dev"
+Requires-Dist: httpx<1.0,>=0.27; extra == "dev"
+Requires-Dist: fastapi<1.0,>=0.111; extra == "dev"
+Requires-Dist: uvicorn[standard]<1.0,>=0.30; extra == "dev"
+Dynamic: license-file
+
+# Syscon TTS
+
+Offline neural text-to-speech for the **PlantStar APU**.
+
+Turns alert text into a voice WAV file with no cloud service, no per-request
+licensing, and no internet connection at runtime. It replaces the APU's
+previous VoiceText integration.
+
+- **Engine:** [Piper](https://github.com/rhasspy/piper) — fast, CPU-only neural
+  TTS. No GPU required.
+- **Interfaces:** a Python library (what the APU uses), a CLI, and an optional
+  HTTP server.
+- **Voices:** six voices across four languages out of the box, configured in a
+  single JSON manifest.
+
+```bash
+pip install syscon-tts
+syscon-tts download-voices      # one-time, needs internet
+syscon-tts doctor               # confirm the install
+```
+
+---
+
+## Platform support
+
+Piper ships **Linux-only** binary wheels. Rather than making that a hard
+install failure everywhere else, the dependency is platform-gated, so the same
+`pip install syscon-tts` works on every platform and gives you what that
+platform can actually do:
+
+| Platform | Synthesis | Replay cached audio | Library / CLI / manifest |
+|---|:--:|:--:|:--:|
+| **Linux x86-64, Python 3.9–3.11** (APU, customer machines) | yes | yes | yes |
+| **Windows, macOS** (developer machines) | no | yes | yes |
+| Linux, Python 3.12+ | no | yes | yes |
+
+Developer machines can therefore install the package, browse voices, run the
+test suite, and exercise the whole alert pipeline against WAVs generated on a
+Linux box — they just can't generate new audio. Attempting to do so raises
+`SynthesisUnavailableError`, which names the platform and explains why.
+
+Run `syscon-tts doctor` any time to see which mode you're in.
+
+> On Linux with Python 3.12+, Piper is skipped rather than failing the install.
+> That is a deliberate trade — `doctor` reports it as a problem, so make it part
+> of provisioning rather than discovering it at the first alert.
+
+---
+
+## Using it from the APU
+
+The only contract the rest of the APU depends on is the side effect: a WAV
+exists at `MEDIA_ROOT/public_alert_sounds/<file_name>.wav`. Everything
+downstream — the websocket push, `/media/` serving, `check_for_file_delete` —
+is agnostic about how the file got there. So the integration is a rewrite of
+`call_tts()` and nothing else.
+
+```python
+from syscon_tts import AlertSynthesizer
+
+# Build once and keep it. Each instance caches loaded voice models in memory;
+# constructing one per alert reintroduces a multi-second cold load every time.
+synthesizer = AlertSynthesizer()
+
+result = synthesizer.ensure(
+    message,
+    file_name=file_name,     # the APU's get_valid_filename(message)[:50]
+    alerts_dir=Path(settings.MEDIA_ROOT, "public_alert_sounds"),
+)
+# result.path    -> the WAV on disk
+# result.cached  -> True if it already existed (no synthesis happened)
+```
+
+`ensure()` is **cache first**: if the WAV is already there it returns
+immediately without touching Piper. Only a cache miss synthesizes.
+
+The package derives file names with the exact algorithm the APU already uses
+(`django.utils.text.get_valid_filename(text)[:50]`), verified byte-for-byte
+including Unicode and the empty-name error cases — so both sides always agree
+on the path. Passing `file_name` explicitly is still recommended, so the APU
+stays the single source of truth.
+
+Full integration recipe, including the `IS_LIVE` gate and voice-ID mapping:
+[DEPLOY.md](DEPLOY.md).
+
+---
+
+## Voice profiles (default set)
+
+| ID | Language | Voice |
+|---|---|---|
+| `en_us_amy` | en_US | Amy — female (default) |
+| `en_us_ryan` | en_US | Ryan — male |
+| `en_gb_alan` | en_GB | Alan — male |
+| `es_es_davefx` | es_ES | DaveFX — male (Spanish) |
+| `fr_fr_siwis` | fr_FR | Siwis — female (French) |
+| `de_de_thorsten` | de_DE | Thorsten — male (German) |
+
+The manifest ships inside the wheel. To use your own catalogue, point
+`SYSCON_TTS_MANIFEST` at a JSON file of the same shape, then run
+`syscon-tts download-voices`. Browse the full ~40-language catalogue at
+<https://huggingface.co/rhasspy/piper-voices>.
+
+---
+
+## CLI
+
+```bash
+syscon-tts doctor                    # platform, Piper, and model status
+syscon-tts download-voices           # fetch all voices (needs internet)
+syscon-tts download-voices en_us_amy # or just one
+syscon-tts list-voices               # catalogue + install state
+
+syscon-tts speak -v en_us_amy -o hello.wav "Hello from PlantStar"
+syscon-tts speak -v de_de_thorsten -f mp3 -o ansage.mp3 --text-file notice.txt
+
+syscon-tts alert "Press 4 cavity pressure exceeded."   # cache-first, as the APU does
+syscon-tts serve --port 5002                           # needs [server] extra
+```
+
+`alert` exits `2` when the audio is absent *and* this machine cannot
+synthesize — distinguishable from `1` for ordinary errors.
+
+---
+
+## Configuration
+
+All optional; every one has a working default.
+
+| Variable | Default | Purpose |
+|---|---|---|
+| `SYSCON_TTS_MANIFEST` | bundled `data/voices.json` | Voice catalogue |
+| `SYSCON_TTS_VOICES_DIR` | platform data dir | Where `.onnx` models live |
+| `SYSCON_TTS_ALERTS_DIR` | platform data dir | Where alert WAVs are written |
+| `SYSCON_TTS_DEFAULT_VOICE` | `en_us_amy` | Voice used when none is given |
+| `SYSCON_TTS_HOST` | `127.0.0.1` | HTTP server bind host |
+| `SYSCON_TTS_PORT` | `5002` | HTTP server bind port |
+| `SYSCON_TTS_MAX_CHARS` | `20000` | Max text length per request |
+
+The platform data dir is `/var/lib/syscon-tts` on Linux when writable
+(otherwise `~/.local/share/syscon-tts`), `%LOCALAPPDATA%\syscon-tts` on
+Windows, and `~/Library/Application Support/syscon-tts` on macOS.
+
+**On the APU**, override the two path variables so audio lands where Django
+already serves it:
+
+```
+SYSCON_TTS_VOICES_DIR=<MEDIA_ROOT>/tts_voices
+SYSCON_TTS_ALERTS_DIR=<MEDIA_ROOT>/public_alert_sounds
+```
+
+`MEDIA_ROOT` is already gitignored and is where every other APU runtime data
+directory lives, so the ~360 MB of voice models stay out of the source tree.
+
+---
+
+## Optional HTTP server
+
+```bash
+pip install 'syscon-tts[server]'
+syscon-tts serve --port 5002
+```
+
+| Method | Path | Description |
+|---|---|---|
+| GET | `/health` | Liveness + whether this host can synthesize |
+| GET | `/voices` | Voice profiles and install state |
+| POST | `/synthesize` | Render text → audio bytes |
+
+```bash
+curl -s -X POST http://localhost:5002/synthesize \
+  -H 'Content-Type: application/json' \
+  -d '{"text":"Machine 4 cycle complete.","voice":"en_us_ryan"}' \
+  -o alert.wav
+```
+
+Status codes: `404` unknown voice · `501` this host can never synthesize
+(no Piper) · `503` voice model not downloaded · `413` text too long ·
+`400` bad input.
+
+The APU does **not** need this — it imports the library in-process, which keeps
+FastAPI and uvicorn out of the Django environment entirely.
+
+---
+
+## Development
+
+```bash
+git clone https://github.com/SYSCON-International/syscon_tts.git
+cd syscon_tts
+python -m venv .venv && . .venv/bin/activate   # Windows: .venv\Scripts\activate
+pip install -e ".[dev]"
+pytest
+```
+
+The full test suite passes on Windows and macOS without Piper — synthesis is
+mocked throughout, so no voice models are needed. See
+[CONTRIBUTING.md](CONTRIBUTING.md).
+
+---
+
+## Documentation
+
+- [DEPLOY.md](DEPLOY.md) — provisioning an APU, the `call_tts()` replacement,
+  air-gapped installs, tuning.
+- [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md) — module layout, design
+  decisions, operations, troubleshooting.
+
+---
+
+## Performance
+
+Piper runs on CPU, so throughput depends on the host. Measure on the target:
+
+```bash
+python scripts/benchmark.py            # per-voice cold load, memory, RTF, latency
+```
+
+It reports the **real-time factor** (RTF = synthesis time ÷ audio seconds) for
+short/medium/long messages. Record results with
+[docs/bench-results-template.md](docs/bench-results-template.md) and use them to
+pick quality tiers — switching a voice to its `low`-quality model in the
+manifest is the main lever on a weak CPU.
+
+Expect roughly 50–120 MB resident per loaded `medium` voice; models stay in
+memory after first use to avoid reload latency.
+
+## License
+
+MIT — see [LICENSE](LICENSE).

syscon_tts-0.0.1/README.md

@@ -0,0 +1,230 @@
+# Syscon TTS
+
+Offline neural text-to-speech for the **PlantStar APU**.
+
+Turns alert text into a voice WAV file with no cloud service, no per-request
+licensing, and no internet connection at runtime. It replaces the APU's
+previous VoiceText integration.
+
+- **Engine:** [Piper](https://github.com/rhasspy/piper) — fast, CPU-only neural
+  TTS. No GPU required.
+- **Interfaces:** a Python library (what the APU uses), a CLI, and an optional
+  HTTP server.
+- **Voices:** six voices across four languages out of the box, configured in a
+  single JSON manifest.
+
+```bash
+pip install syscon-tts
+syscon-tts download-voices      # one-time, needs internet
+syscon-tts doctor               # confirm the install
+```
+
+---
+
+## Platform support
+
+Piper ships **Linux-only** binary wheels. Rather than making that a hard
+install failure everywhere else, the dependency is platform-gated, so the same
+`pip install syscon-tts` works on every platform and gives you what that
+platform can actually do:
+
+| Platform | Synthesis | Replay cached audio | Library / CLI / manifest |
+|---|:--:|:--:|:--:|
+| **Linux x86-64, Python 3.9–3.11** (APU, customer machines) | yes | yes | yes |
+| **Windows, macOS** (developer machines) | no | yes | yes |
+| Linux, Python 3.12+ | no | yes | yes |
+
+Developer machines can therefore install the package, browse voices, run the
+test suite, and exercise the whole alert pipeline against WAVs generated on a
+Linux box — they just can't generate new audio. Attempting to do so raises
+`SynthesisUnavailableError`, which names the platform and explains why.
+
+Run `syscon-tts doctor` any time to see which mode you're in.
+
+> On Linux with Python 3.12+, Piper is skipped rather than failing the install.
+> That is a deliberate trade — `doctor` reports it as a problem, so make it part
+> of provisioning rather than discovering it at the first alert.
+
+---
+
+## Using it from the APU
+
+The only contract the rest of the APU depends on is the side effect: a WAV
+exists at `MEDIA_ROOT/public_alert_sounds/<file_name>.wav`. Everything
+downstream — the websocket push, `/media/` serving, `check_for_file_delete` —
+is agnostic about how the file got there. So the integration is a rewrite of
+`call_tts()` and nothing else.
+
+```python
+from syscon_tts import AlertSynthesizer
+
+# Build once and keep it. Each instance caches loaded voice models in memory;
+# constructing one per alert reintroduces a multi-second cold load every time.
+synthesizer = AlertSynthesizer()
+
+result = synthesizer.ensure(
+    message,
+    file_name=file_name,     # the APU's get_valid_filename(message)[:50]
+    alerts_dir=Path(settings.MEDIA_ROOT, "public_alert_sounds"),
+)
+# result.path    -> the WAV on disk
+# result.cached  -> True if it already existed (no synthesis happened)
+```
+
+`ensure()` is **cache first**: if the WAV is already there it returns
+immediately without touching Piper. Only a cache miss synthesizes.
+
+The package derives file names with the exact algorithm the APU already uses
+(`django.utils.text.get_valid_filename(text)[:50]`), verified byte-for-byte
+including Unicode and the empty-name error cases — so both sides always agree
+on the path. Passing `file_name` explicitly is still recommended, so the APU
+stays the single source of truth.
+
+Full integration recipe, including the `IS_LIVE` gate and voice-ID mapping:
+[DEPLOY.md](DEPLOY.md).
+
+---
+
+## Voice profiles (default set)
+
+| ID | Language | Voice |
+|---|---|---|
+| `en_us_amy` | en_US | Amy — female (default) |
+| `en_us_ryan` | en_US | Ryan — male |
+| `en_gb_alan` | en_GB | Alan — male |
+| `es_es_davefx` | es_ES | DaveFX — male (Spanish) |
+| `fr_fr_siwis` | fr_FR | Siwis — female (French) |
+| `de_de_thorsten` | de_DE | Thorsten — male (German) |
+
+The manifest ships inside the wheel. To use your own catalogue, point
+`SYSCON_TTS_MANIFEST` at a JSON file of the same shape, then run
+`syscon-tts download-voices`. Browse the full ~40-language catalogue at
+<https://huggingface.co/rhasspy/piper-voices>.
+
+---
+
+## CLI
+
+```bash
+syscon-tts doctor                    # platform, Piper, and model status
+syscon-tts download-voices           # fetch all voices (needs internet)
+syscon-tts download-voices en_us_amy # or just one
+syscon-tts list-voices               # catalogue + install state
+
+syscon-tts speak -v en_us_amy -o hello.wav "Hello from PlantStar"
+syscon-tts speak -v de_de_thorsten -f mp3 -o ansage.mp3 --text-file notice.txt
+
+syscon-tts alert "Press 4 cavity pressure exceeded."   # cache-first, as the APU does
+syscon-tts serve --port 5002                           # needs [server] extra
+```
+
+`alert` exits `2` when the audio is absent *and* this machine cannot
+synthesize — distinguishable from `1` for ordinary errors.
+
+---
+
+## Configuration
+
+All optional; every one has a working default.
+
+| Variable | Default | Purpose |
+|---|---|---|
+| `SYSCON_TTS_MANIFEST` | bundled `data/voices.json` | Voice catalogue |
+| `SYSCON_TTS_VOICES_DIR` | platform data dir | Where `.onnx` models live |
+| `SYSCON_TTS_ALERTS_DIR` | platform data dir | Where alert WAVs are written |
+| `SYSCON_TTS_DEFAULT_VOICE` | `en_us_amy` | Voice used when none is given |
+| `SYSCON_TTS_HOST` | `127.0.0.1` | HTTP server bind host |
+| `SYSCON_TTS_PORT` | `5002` | HTTP server bind port |
+| `SYSCON_TTS_MAX_CHARS` | `20000` | Max text length per request |
+
+The platform data dir is `/var/lib/syscon-tts` on Linux when writable
+(otherwise `~/.local/share/syscon-tts`), `%LOCALAPPDATA%\syscon-tts` on
+Windows, and `~/Library/Application Support/syscon-tts` on macOS.
+
+**On the APU**, override the two path variables so audio lands where Django
+already serves it:
+
+```
+SYSCON_TTS_VOICES_DIR=<MEDIA_ROOT>/tts_voices
+SYSCON_TTS_ALERTS_DIR=<MEDIA_ROOT>/public_alert_sounds
+```
+
+`MEDIA_ROOT` is already gitignored and is where every other APU runtime data
+directory lives, so the ~360 MB of voice models stay out of the source tree.
+
+---
+
+## Optional HTTP server
+
+```bash
+pip install 'syscon-tts[server]'
+syscon-tts serve --port 5002
+```
+
+| Method | Path | Description |
+|---|---|---|
+| GET | `/health` | Liveness + whether this host can synthesize |
+| GET | `/voices` | Voice profiles and install state |
+| POST | `/synthesize` | Render text → audio bytes |
+
+```bash
+curl -s -X POST http://localhost:5002/synthesize \
+  -H 'Content-Type: application/json' \
+  -d '{"text":"Machine 4 cycle complete.","voice":"en_us_ryan"}' \
+  -o alert.wav
+```
+
+Status codes: `404` unknown voice · `501` this host can never synthesize
+(no Piper) · `503` voice model not downloaded · `413` text too long ·
+`400` bad input.
+
+The APU does **not** need this — it imports the library in-process, which keeps
+FastAPI and uvicorn out of the Django environment entirely.
+
+---
+
+## Development
+
+```bash
+git clone https://github.com/SYSCON-International/syscon_tts.git
+cd syscon_tts
+python -m venv .venv && . .venv/bin/activate   # Windows: .venv\Scripts\activate
+pip install -e ".[dev]"
+pytest
+```
+
+The full test suite passes on Windows and macOS without Piper — synthesis is
+mocked throughout, so no voice models are needed. See
+[CONTRIBUTING.md](CONTRIBUTING.md).
+
+---
+
+## Documentation
+
+- [DEPLOY.md](DEPLOY.md) — provisioning an APU, the `call_tts()` replacement,
+  air-gapped installs, tuning.
+- [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md) — module layout, design
+  decisions, operations, troubleshooting.
+
+---
+
+## Performance
+
+Piper runs on CPU, so throughput depends on the host. Measure on the target:
+
+```bash
+python scripts/benchmark.py            # per-voice cold load, memory, RTF, latency
+```
+
+It reports the **real-time factor** (RTF = synthesis time ÷ audio seconds) for
+short/medium/long messages. Record results with
+[docs/bench-results-template.md](docs/bench-results-template.md) and use them to
+pick quality tiers — switching a voice to its `low`-quality model in the
+manifest is the main lever on a weak CPU.
+
+Expect roughly 50–120 MB resident per loaded `medium` voice; models stay in
+memory after first use to avoid reload latency.
+
+## License
+
+MIT — see [LICENSE](LICENSE).

syscon_tts-0.0.1/pyproject.toml

@@ -0,0 +1,76 @@
+[build-system]
+requires = ["setuptools>=77"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "syscon-tts"
+version = "0.0.1"
+description = "Offline neural text-to-speech for the PlantStar APU"
+readme = "README.md"
+requires-python = ">=3.9"
+license = "MIT"
+license-files = ["LICENSE"]
+# Split across two entries so the metadata renders as separate `Author` and
+# `Author-email` fields, matching django-search-filter-sort. A single entry
+# with both keys would collapse into `Author-email: Name <addr>` instead.
+authors = [
+    { name = "Keith Hostetler" },
+    { email = "robh@syscon-intl.com" },
+]
+keywords = ["tts", "text-to-speech", "piper", "offline", "plantstar", "apu"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Intended Audience :: Manufacturing",
+    "Operating System :: POSIX :: Linux",
+    "Operating System :: MacOS :: MacOS X",
+    "Operating System :: Microsoft :: Windows",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.9",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Topic :: Multimedia :: Sound/Audio :: Speech",
+]
+
+# Piper ships Linux-only binary wheels (piper-phonemize has no Windows or
+# macOS distribution, and none for Python 3.12+). The marker keeps
+# `pip install syscon-tts` working everywhere: Linux hosts get a synthesis
+# engine, other platforms get a package that can still resolve, inspect, and
+# replay previously generated audio. Run `syscon-tts doctor` after installing
+# to confirm which mode you landed in.
+dependencies = [
+    "piper-tts==1.2.0; sys_platform == 'linux' and python_version < '3.12'",
+]
+
+[project.optional-dependencies]
+# Force the synthesis engine regardless of the marker above. Fails loudly on a
+# platform Piper cannot support, which is what you want when the install is
+# meant to produce audio.
+piper = ["piper-tts==1.2.0"]
+# The optional HTTP server. Not needed by the APU, which imports the library.
+server = ["fastapi>=0.111,<1.0", "uvicorn[standard]>=0.30,<1.0"]
+dev = [
+    "pytest>=8.2,<9",
+    "httpx>=0.27,<1.0",
+    "fastapi>=0.111,<1.0",
+    "uvicorn[standard]>=0.30,<1.0",
+]
+
+[project.urls]
+Homepage = "https://github.com/SYSCON-International/syscon_tts"
+Source = "https://github.com/SYSCON-International/syscon_tts"
+Issues = "https://github.com/SYSCON-International/syscon_tts/issues"
+
+[project.scripts]
+syscon-tts = "syscon_tts.cli:main"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.setuptools.package-data]
+# The voice manifest travels inside the wheel; the models it points at are
+# downloaded at provisioning time (they are far too large to ship).
+syscon_tts = ["data/*.json"]
+
+[tool.pytest.ini_options]
+pythonpath = ["src"]
+testpaths = ["tests"]

syscon_tts-0.0.1/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

syscon_tts-0.0.1/src/syscon_tts/__init__.py

@@ -0,0 +1,65 @@
+"""Syscon TTS - offline text-to-speech for the PlantStar APU.
+
+Wraps the Piper neural TTS engine (CPU-only, fully offline) behind a small
+library, a command-line tool, and an optional HTTP server.
+
+Piper only installs on Linux, which is where every APU runs. Importing this
+package is nonetheless safe on Windows and macOS: the Piper import is deferred
+until synthesis is attempted, so developers on those platforms can install the
+package, inspect voices, and replay audio generated elsewhere. Attempting to
+generate new audio there raises
+:class:`~syscon_tts.engine.SynthesisUnavailableError`.
+
+Typical APU usage -- resolve alert text to a WAV, synthesizing only on a cache
+miss::
+
+    from syscon_tts import AlertSynthesizer
+
+    synth = AlertSynthesizer()          # build once, keep it: caches models
+    result = synth.ensure("Press 4 cavity pressure exceeded.")
+    print(result.path, result.cached)
+"""
+
+from .alerts import (
+    AlertAudio,
+    AlertSynthesizer,
+    InvalidAlertNameError,
+    ensure_alert_wav,
+    sanitize_file_name,
+)
+from .config import Settings, load_settings
+from .engine import (
+    SynthesisError,
+    SynthesisUnavailableError,
+    TTSEngine,
+    piper_available,
+)
+from .voices import (
+    UnknownVoiceError,
+    VoiceError,
+    VoiceNotInstalledError,
+    VoiceProfile,
+    VoiceRegistry,
+)
+
+__version__ = "0.0.1"
+
+__all__ = [
+    "AlertAudio",
+    "AlertSynthesizer",
+    "InvalidAlertNameError",
+    "Settings",
+    "SynthesisError",
+    "SynthesisUnavailableError",
+    "TTSEngine",
+    "UnknownVoiceError",
+    "VoiceError",
+    "VoiceNotInstalledError",
+    "VoiceProfile",
+    "VoiceRegistry",
+    "__version__",
+    "ensure_alert_wav",
+    "load_settings",
+    "piper_available",
+    "sanitize_file_name",
+]