Start work on getting rid of piper code and only using the model data

The inference code if a few hundred lines anyway, piper is unmaintained
and forked and the fork depends on the gigantic python ONNXRuntime
bindings. Just write the few hundred lines myself.

setup/build_environment.py

@@ -153,6 +153,8 @@ uchardet_inc_dirs, uchardet_lib_dirs, uchardet_libs = [], [], ['uchardet']
 
 openssl_inc_dirs, openssl_lib_dirs = [], []
 
+piper_inc_dirs, piper_lib_dirs, piper_libs = [], [], []
+
 ICU = sw = ''
 
 if iswindows:
@@ -211,6 +213,10 @@ else:
     uchardet_inc_dirs = pkgconfig_include_dirs('uchardet', '', '/usr/include/uchardet')
     uchardet_lib_dirs = pkgconfig_lib_dirs('uchardet', '', '/usr/lib')
     uchardet_libs = pkgconfig_libs('uchardet', '', '')
+    piper_inc_dirs = pkgconfig_include_dirs('espeak-ng', '', '/usr/include') + pkgconfig_include_dirs(
+            'libonnxruntime', '', '/usr/include/onnxruntime')
+    piper_lib_dirs = pkgconfig_lib_dirs('espeak-ng', '', '/usr/lib') + pkgconfig_lib_dirs('libonnxruntime', '', '/usr/lib')
+    piper_libs = pkgconfig_libs('espeak-ng', '', 'espeak-ng') + pkgconfig_libs('libonnxruntime', '', 'onnxruntime')
     for x in ('libavcodec', 'libavformat', 'libavdevice', 'libavfilter', 'libavutil', 'libpostproc', 'libswresample', 'libswscale'):
         for inc in pkgconfig_include_dirs(x, '', '/usr/include'):
             if inc and inc not in ffmpeg_inc_dirs:

setup/extensions.json

@@ -134,6 +134,14 @@
         "error": "!podofo_error",
 		"needs_c++": "17"
     },
+    {
+        "name": "piper",
+        "sources": "calibre/utils/tts/piper.cpp",
+		"needs_c++": "17",
+        "libraries": "!piper_libs",
+        "lib_dirs": "!piper_lib_dirs",
+        "inc_dirs": "!piper_inc_dirs"
+    },
     {
         "name": "html_as_json",
         "sources": "calibre/srv/html_as_json.cpp",

src/calibre/constants.py

@@ -258,6 +258,7 @@ class ExtensionsImporter:
             'rcc_backend',
             'icu',
             'speedup',
+            'piper',
             'html_as_json',
             'fast_css_transform',
             'fast_html_entities',

src/calibre/utils/tts/piper.cpp

@@ -0,0 +1,39 @@
+/*
+ * piper.cpp
+ * Copyright (C) 2025 Kovid Goyal <kovid at kovidgoyal.net>
+ *
+ * Distributed under terms of the GPL3 license.
+ */
+#define PY_SSIZE_T_CLEAN
+
+#include <Python.h>
+#include <espeak-ng/speak_lib.h>
+
+static PyObject*
+phonemize(PyObject *self, PyObject *args) {
+    Py_RETURN_NONE;
+}
+
+// Boilerplate {{{
+static char doc[] = "Text to speech using the Piper TTS models";
+static PyMethodDef methods[] = {
+    {"phonemize", (PyCFunction)phonemize, METH_VARARGS,
+     "Convert the specified text into espeak-ng phonemes"
+    },
+    {NULL}  /* Sentinel */
+};
+static int
+exec_module(PyObject *mod) { return 0; }
+
+static PyModuleDef_Slot slots[] = { {Py_mod_exec, (void*)exec_module}, {0, NULL} };
+
+static struct PyModuleDef module_def = {PyModuleDef_HEAD_INIT};
+
+CALIBRE_MODINIT_FUNC PyInit_piper(void) {
+	module_def.m_name = "piper";
+	module_def.m_slots = slots;
+	module_def.m_doc = doc;
+	module_def.m_methods = methods;
+	return PyModuleDef_Init(&module_def);
+}
+// }}}

src/calibre/utils/tts/piper.py

@@ -0,0 +1,6 @@
+#!/usr/bin/env python
+# License: GPLv3 Copyright: 2025, Kovid Goyal <kovid at kovidgoyal.net>
+
+import calibre_extensions.piper as piper
+
+piper