Updated docs + added TTS backends

README.md

@@ -18,6 +18,7 @@ The name is a coined Japanese compound from `採` as in gathering/collecting and
 - [Anki](https://apps.ankiweb.net/) with [AnkiConnect](https://github.com/amikey/anki-connect)
 - `ffmpeg`
 - Python dependencies from `requirements.txt`
+- Optional extra TTS backend tools: `piper`, `espeak-ng`, and `kokoro-onnx`.
 - spaCy models for word mining:
 
 ```shell
@@ -35,6 +36,62 @@ pip install -r requirements.txt
 sudo dnf install ffmpeg
 ```
 
+### Optional TTS Backends
+
+The default `edge-tts` backend is installed by `requirements.txt`. Install only
+the optional pieces you plan to test:
+
+```shell
+# Python-backed optional engines: piper, kokoro.
+pip install -r requirements-tts.txt
+
+# System package for espeak-ng.
+sudo dnf install espeak-ng
+```
+
+Other package-manager names:
+
+```shell
+sudo apt-get install espeak-ng
+sudo pacman -S espeak-ng
+```
+
+Backend notes:
+
+- `edge-tts`: installed by `pip install edge-tts`; no API key, but it uses
+  Microsoft Edge's online TTS service.
+- `gtts`: installed by `requirements.txt`; no API key, but it uses Google's
+  online TTS service through `gtts-cli`.
+- `piper`: installed by `pip install piper-tts`; you still need a compatible
+  `.onnx` voice model, usually with its matching `.onnx.json` config file.
+- `espeak-ng`: installed through your OS package manager, not pip.
+- `kokoro`: installed by `pip install kokoro-onnx soundfile`; you still need
+  `kokoro-v1.0.onnx` and `voices-v1.0.bin`, plus any language-specific G2P
+  setup required by your Kokoro release.
+
+Example model downloads for the README smoke tests:
+
+```shell
+mkdir -p ~/.local/share/saiki/models
+
+# Piper Spanish voice model plus matching config.
+wget -O ~/.local/share/saiki/models/es_ES-davefx-medium.onnx \
+  https://huggingface.co/rhasspy/piper-voices/resolve/main/es/es_ES/davefx/medium/es_ES-davefx-medium.onnx
+wget -O ~/.local/share/saiki/models/es_ES-davefx-medium.onnx.json \
+  https://huggingface.co/rhasspy/piper-voices/resolve/main/es/es_ES/davefx/medium/es_ES-davefx-medium.onnx.json
+
+# Kokoro ONNX model plus voices bundle.
+wget -O ~/.local/share/saiki/models/kokoro-v1.0.onnx \
+  https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files-v1.0/kokoro-v1.0.onnx
+wget -O ~/.local/share/saiki/models/voices-v1.0.bin \
+  https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files-v1.0/voices-v1.0.bin
+```
+
+Saiki's default `tts_model_dir` is `~/.local/share/saiki/models`. Relative
+model paths such as `es_ES-davefx-medium.onnx` are resolved under that
+directory. You can override it in YAML with `tts_model_dir` or for one command
+with `--tts-model-dir`.
+
 ## Configuration
 
 Defaults are built in, but you can override them with YAML:
@@ -57,6 +114,7 @@ media_dir: ~/.var/app/net.ankiweb.Anki/data/Anki2/User 1/collection.media
 audio_output_root: ~/Languages/Anki/anki-audio
 word_output_root: ~/Languages/Anki/anki-words
 sentence_dir: ~/Languages/Anki
+tts_model_dir: ~/.local/share/saiki/models
 note_model: Basic
 fields:
   front: Front
@@ -65,8 +123,8 @@ languages:
   jp:
     name: japanese
     transcript_code: ja
-    tts_code: ja
+    tts_backend: edge-tts
-    tts_tld: com
+    tts_voice: ja-JP-NanamiNeural
     tts_tempo: 1.35
     decks: ["日本語"]
     field: Back
@@ -75,8 +133,8 @@ languages:
   es:
     name: spanish
     transcript_code: es
-    tts_code: es
+    tts_backend: edge-tts
-    tts_tld: es
+    tts_voice: es-ES-ElviraNeural
     tts_tempo: 1.25
     decks: ["Español"]
     field: Back
@@ -174,12 +232,59 @@ Generate TTS audio and add sentence cards to Anki.
 ./saiki.py import es
 ./saiki.py import jp ~/Languages/Anki/sentences_jp.txt
 ./saiki.py import es youtube.tsv --tags youtube,manual
+./saiki.py import es --tts-voice es-MX-DaliaNeural
 ```
 
 The importer accepts plain text sentence files and TSV/CSV files with a
 `sentence` column. `text-to-speech` is always added as a tag. If `--tags` is not
 provided, `AI-generated` is added.
 
+TTS is configured per language with `tts_backend`. Supported backends are:
+
+- `edge-tts`: default backend using Microsoft Edge neural voices; configure
+  `tts_voice`.
+- `gtts`: free backend using `gtts-cli`; configure `tts_code` and
+  `tts_tld`.
+- `piper`: local/offline neural TTS; configure `tts_model` with a model path.
+  The stock Piper catalog includes Spanish voices, but not Japanese.
+- `espeak-ng`: local/offline lightweight TTS; configure `tts_voice`. Spanish is
+  supported; Japanese is documented as kana-only and is not recommended for
+  normal Japanese sentence cards.
+- `kokoro`: local/offline neural TTS; configure `tts_model`, `tts_voices`,
+  `tts_voice`, and `tts_code`; some Japanese setups also need
+  `tts_vocab_config`. Kokoro lists Japanese and Spanish voices, but upstream
+  notes that non-English quality can be thin.
+
+You can override backend settings for one import:
+
+```shell
+./saiki.py import jp sentences_jp.txt \
+  --tts-backend edge-tts \
+  --tts-voice ja-JP-KeitaNeural
+```
+
+Voice-listing helpers:
+
+```shell
+./saiki.py tts-voices jp
+./saiki.py tts-voices es --backend edge-tts
+```
+
+Test a TTS backend without creating Anki cards:
+
+```shell
+./saiki.py tts-test es --out /tmp/saiki_edge_default_es.mp3
+./saiki.py tts-test jp --tts-backend edge-tts --tts-voice ja-JP-NanamiNeural --out /tmp/saiki_edge_jp.mp3
+./saiki.py tts-test es --tts-backend edge-tts --tts-voice es-ES-ElviraNeural --out /tmp/saiki_edge_es.mp3
+./saiki.py tts-test es --tts-backend gtts --tts-code es --tts-tld es --out /tmp/saiki_gtts_es.mp3
+./saiki.py tts-test es --tts-backend piper --tts-model es_ES-davefx-medium.onnx --tts-config es_ES-davefx-medium.onnx.json --out /tmp/saiki_piper_es.mp3
+./saiki.py tts-test es --tts-backend espeak-ng --tts-voice es --out /tmp/saiki_espeak_es.mp3
+./saiki.py tts-test es --tts-backend kokoro --tts-model kokoro-v1.0.onnx --tts-voices voices-v1.0.bin --tts-voice ef_dora --out /tmp/saiki_kokoro_es.mp3
+```
+
+For `kokoro`, put `tts_model`, `tts_voices`, and any needed `tts_vocab_config`
+in your config file rather than typing every path each time.
+
 ### Known/New Words
 
 Compare any generated word list against an existing known list:

examples/config.yaml

@@ -11,6 +11,7 @@ media_dir: ~/.var/app/net.ankiweb.Anki/data/Anki2/User 1/collection.media
 audio_output_root: ~/Languages/Anki/anki-audio
 word_output_root: ~/Languages/Anki/anki-words
 sentence_dir: ~/Languages/Anki
+tts_model_dir: ~/.local/share/saiki/models
 note_model: Basic
 
 fields:
@@ -21,8 +22,6 @@ languages:
   jp:
     name: japanese
     transcript_code: ja
-    tts_code: ja
-    tts_tld: com
     tts_tempo: 1.35
     decks:
       - 日本語
@@ -30,14 +29,69 @@ languages:
     word_model: ja_core_news_lg
     sentence_file: sentences_jp.txt
 
+    # --- TTS backend (pick one) ---
+
+    # edge-tts (default): Microsoft Edge neural TTS. Requires: pip install edge-tts
+    tts_backend: edge-tts
+    tts_voice: ja-JP-NanamiNeural   # or ja-JP-KeitaNeural for male
+
+    # gtts: free Google TTS via gtts-cli. Requires: pip install gtts
+    # tts_backend: gtts
+    # tts_code: ja
+    # tts_tld: com
+
+    # piper: no stock Japanese voice is listed in Piper's official voice catalog.
+    # Only use this backend for JP if you have your own compatible Japanese model.
+
+    # espeak-ng: Japanese support is kana-only in the upstream docs, so this is
+    # not a good fit for normal Japanese sentences that include kanji.
+
+    # kokoro: offline neural TTS. Requires: pip install kokoro-onnx soundfile
+    # plus the Kokoro model/voice files and Japanese G2P dependencies for your
+    # installed Kokoro release. Kokoro lists Japanese voices, but its docs warn
+    # that non-English quality can be thin.
+    # tts_backend: kokoro
+    # tts_model: kokoro-v1.0.onnx
+    # tts_voices: voices-v1.0.bin
+    # tts_vocab_config: kokoro-ja-config.json
+    # tts_voice: jf_alpha   # Japanese female; jm_kumo for male
+    # tts_code: ja
+
   es:
     name: spanish
     transcript_code: es
-    tts_code: es
-    tts_tld: es
     tts_tempo: 1.25
     decks:
       - Español
     field: Back
     word_model: es_core_news_sm
     sentence_file: sentences_es.txt
+
+    # --- TTS backend (pick one) ---
+
+    # edge-tts (default): Microsoft Edge neural TTS. Requires: pip install edge-tts
+    tts_backend: edge-tts
+    tts_voice: es-ES-ElviraNeural   # or es-MX-DaliaNeural for Mexican Spanish
+
+    # gtts: free Google TTS via gtts-cli. Requires: pip install gtts
+    # tts_backend: gtts
+    # tts_code: es
+    # tts_tld: es
+
+    # piper: offline neural TTS. Requires: piper binary + model download
+    # tts_backend: piper
+    # tts_model: es_ES-davefx-medium.onnx
+    # tts_config: es_ES-davefx-medium.onnx.json
+
+    # espeak-ng: offline, lightweight, robotic quality. Requires: espeak-ng package
+    # tts_backend: espeak-ng
+    # tts_voice: es
+
+    # kokoro: offline neural TTS. Requires: pip install kokoro-onnx soundfile
+    # plus the Kokoro model/voice files. Kokoro lists Spanish voices, but its
+    # docs warn that non-English quality can be thin.
+    # tts_backend: kokoro
+    # tts_model: kokoro-v1.0.onnx
+    # tts_voices: voices-v1.0.bin
+    # tts_voice: ef_dora   # Spanish female
+    # tts_code: es

requirements-tts.txt

@@ -0,0 +1,5 @@
+# Optional free TTS backends.
+# Install this only if you want to test/use non-default TTS engines.
+piper-tts
+kokoro-onnx
+soundfile

requirements.txt

@@ -4,5 +4,6 @@ spacy
 youtube-transcript-api
 fugashi[unidic-lite]
 gTTS
+edge-tts
 pyyaml
 genanki

saiki.py

@@ -6,6 +6,8 @@ from pathlib import Path
 
 SRC_DIR = Path(__file__).resolve().parent / "src"
 if str(SRC_DIR) not in sys.path:
+    # Let the repository checkout run directly without requiring an editable
+    # install first. Installed packages will still resolve normally.
     sys.path.insert(0, str(SRC_DIR))
 
 from saiki.cli import main

src/saiki/ankiconnect.py

@@ -6,6 +6,13 @@ import requests
 
 
 def anki_request(action: str, url: str = "http://localhost:8765", **params):
+    """Send one JSON-RPC style request to AnkiConnect.
+
+    AnkiConnect exposes all operations as an ``action`` plus a ``params``
+    object. This helper centralizes the protocol version, timeout, HTTP error
+    handling, and conversion of AnkiConnect's ``error`` field into a Python
+    exception.
+    """
     resp = requests.post(
         url,
         json={"action": action, "version": 6, "params": params},
@@ -16,4 +23,3 @@ def anki_request(action: str, url: str = "http://localhost:8765", **params):
     if data.get("error") is not None:
         raise RuntimeError(f"AnkiConnect error for {action}: {data['error']}")
     return data["result"]
-

src/saiki/audio.py

@@ -16,6 +16,13 @@ AUDIO_EXTS = (".mp3", ".wav", ".ogg", ".m4a", ".flac")
 
 
 def resolve_media_paths(media_dir: str, out_dir: str, media_name: str) -> tuple[str, str] | None:
+    """Return safe source and destination paths for one Anki media filename.
+
+    Anki stores audio references as media names, not arbitrary filesystem
+    paths. Absolute paths and parent-directory traversal are rejected so a
+    malformed card cannot make the export read or write outside the configured
+    media/output directories.
+    """
     normalized = os.path.normpath(media_name)
     if os.path.isabs(normalized) or normalized.startswith(".."):
         return None
@@ -23,6 +30,7 @@ def resolve_media_paths(media_dir: str, out_dir: str, media_name: str) -> tuple[
 
 
 def build_playlist(out_dir: str, language: str) -> str:
+    """Write an M3U playlist containing exported audio files for a language."""
     m3u_path = os.path.join(out_dir, f"{language}.m3u")
     concat_name = f"{language}_concat.mp3"
     files: list[str] = []
@@ -42,6 +50,7 @@ def build_playlist(out_dir: str, language: str) -> str:
 
 
 def concat_audio_from_m3u(out_dir: str, m3u_path: str, out_path: str) -> None:
+    """Concatenate playlist entries into a single MP3 with ffmpeg."""
     if shutil.which("ffmpeg") is None:
         raise RuntimeError("ffmpeg not found in PATH. Install ffmpeg to use --concat.")
 
@@ -59,6 +68,8 @@ def concat_audio_from_m3u(out_dir: str, m3u_path: str, out_path: str) -> None:
     with tempfile.NamedTemporaryFile("w", delete=False, encoding="utf-8") as tmp:
         concat_list_path = tmp.name
         for path in abs_files:
+            # ffmpeg's concat demuxer uses single-quoted paths. Escape literal
+            # apostrophes so media filenames from Anki remain valid entries.
             tmp.write(f"file '{path.replace(chr(39), chr(39) + chr(92) + chr(39) + chr(39))}'\n")
 
     cmd = [
@@ -83,6 +94,13 @@ def extract_audio(
     concat: bool = False,
     request: Callable = anki_request,
 ) -> dict[str, object]:
+    """Copy audio from configured Anki decks and build a playlist.
+
+    The return value is intentionally CLI-friendly: it reports the number of
+    copied files, the playlist path, the output directory, and the optional
+    concatenated MP3 path. ``request`` is injectable so tests can exercise the
+    workflow without a running Anki instance.
+    """
     language = config.language_name(lang)
     selected_decks = config.decks_for(lang)
     if not selected_decks:
@@ -123,4 +141,3 @@ def extract_audio(
         concat_path = os.path.join(out_dir, f"{language}_concat.mp3")
         concat_audio_from_m3u(out_dir, m3u_path, concat_path)
     return {"copied": len(copied), "playlist": m3u_path, "outdir": out_dir, "concat": concat_path}
-

src/saiki/cli.py

@@ -7,17 +7,67 @@ import sys
 
 from .audio import extract_audio
 from .config import Config, language_choices, load_config
-from .importer import import_sentences
+from .importer import (
+    format_tts_error,
+    import_sentences,
+    list_tts_voices,
+    supported_tts_backends,
+    synthesize_tts_sample,
+)
 from .words import compare_word_files, extract_words
 from .youtube import run_youtube
 
 
 def add_config_arg(parser: argparse.ArgumentParser) -> None:
+    """Attach the shared ``--config`` option to a parser."""
     parser.add_argument("--config", help="Path to YAML config file.")
 
 
+def add_tts_override_args(parser: argparse.ArgumentParser, tts_backends: list[str]) -> None:
+    """Attach per-command TTS override flags.
+
+    These options intentionally mirror config keys so command-line overrides
+    can be collected mechanically and merged over the selected language.
+    """
+    parser.add_argument("--tts-backend", choices=tts_backends, help="Override the configured TTS backend.")
+    parser.add_argument("--tts-voice", help="Override the configured backend voice.")
+    parser.add_argument("--tts-voices", help="Override the configured backend voice bundle path.")
+    parser.add_argument("--tts-model", help="Override the configured backend model or local model path.")
+    parser.add_argument("--tts-model-dir", help="Override the directory used for relative TTS model paths.")
+    parser.add_argument("--tts-config", help="Override the configured backend model config path.")
+    parser.add_argument("--tts-vocab-config", help="Override the configured backend vocab config path.")
+    parser.add_argument("--tts-code", help="Override the configured backend language code.")
+    parser.add_argument("--tts-tld", help="Override the configured gTTS top-level domain.")
+    parser.add_argument("--tts-tempo", type=float, help="Override the post-processing tempo multiplier.")
+    parser.add_argument("--tts-speed", type=float, help="Override backend-native speech speed when supported.")
+
+
+def collect_tts_overrides(args: argparse.Namespace) -> dict[str, object]:
+    """Collect TTS override attributes from an argparse namespace."""
+    return {
+        "tts_backend": getattr(args, "tts_backend", None),
+        "tts_voice": getattr(args, "tts_voice", None),
+        "tts_voices": getattr(args, "tts_voices", None),
+        "tts_model": getattr(args, "tts_model", None),
+        "tts_model_dir": getattr(args, "tts_model_dir", None),
+        "tts_config": getattr(args, "tts_config", None),
+        "tts_vocab_config": getattr(args, "tts_vocab_config", None),
+        "tts_code": getattr(args, "tts_code", None),
+        "tts_tld": getattr(args, "tts_tld", None),
+        "tts_tempo": getattr(args, "tts_tempo", None),
+        "tts_speed": getattr(args, "tts_speed", None),
+    }
+
+
 def build_parser(config: Config | None = None) -> argparse.ArgumentParser:
+    """Build the full CLI parser.
+
+    Passing a loaded config lets argparse choices reflect user-defined language
+    codes. When no config is supplied, defaults are loaded so the parser remains
+    usable in tests and help-generation contexts.
+    """
     choices = language_choices(config or load_config())
+    tts_backends = supported_tts_backends()
     parser = argparse.ArgumentParser(description="Saiki: sentence mining and listening tools for Anki.")
     add_config_arg(parser)
     sub = parser.add_subparsers(dest="command", required=True)
@@ -61,11 +111,25 @@ def build_parser(config: Config | None = None) -> argparse.ArgumentParser:
     importer.add_argument("lang", choices=choices)
     importer.add_argument("sentence_file", nargs="?")
     importer.add_argument("--tags", help="Comma-separated tags. text-to-speech is always included.")
+    add_tts_override_args(importer, tts_backends)
+
+    test_tts = sub.add_parser("tts-test", help="Synthesize one TTS sample without importing into Anki.")
+    test_tts.add_argument("lang", choices=choices)
+    test_tts.add_argument("text", nargs="?")
+    test_tts.add_argument("--out", help="Output MP3 path. Defaults to ./tts_test_<lang>_<backend>.mp3.")
+    add_tts_override_args(test_tts, tts_backends)
+
+    voices = sub.add_parser("tts-voices", help="List voices or voice-listing hints for a TTS backend.")
+    voices.add_argument("lang", nargs="?", choices=choices)
+    voices.add_argument("--backend", choices=tts_backends, help="Backend to list instead of the language default.")
 
     return parser
 
 
 def main(argv: list[str] | None = None) -> int:
+    """Run the CLI and return a process exit status."""
+    # Parse --config first so subcommand language choices can come from the
+    # user's config file instead of only the built-in defaults.
     pre = argparse.ArgumentParser(add_help=False)
     add_config_arg(pre)
     known, _ = pre.parse_known_args(argv)
@@ -114,10 +178,27 @@ def main(argv: list[str] | None = None) -> int:
         return 0
 
     if args.command == "import":
-        result = import_sentences(config, args.lang, args.sentence_file, args.tags)
+        tts_overrides = collect_tts_overrides(args)
+        result = import_sentences(config, args.lang, args.sentence_file, args.tags, tts_overrides=tts_overrides)
         print(f"Done. Added {result.added}/{result.processed} cards. Failed: {result.failed}")
+        for error in result.errors:
+            print(f"Error: {error}", file=sys.stderr)
         return 0 if result.failed == 0 else 1
 
+    if args.command == "tts-test":
+        try:
+            output = synthesize_tts_sample(config, args.lang, args.text, args.out, collect_tts_overrides(args))
+            print(f"Wrote TTS sample: {output}")
+            return 0
+        except Exception as exc:
+            print(f"Error: {format_tts_error(exc)}", file=sys.stderr)
+            return 1
+
+    if args.command == "tts-voices":
+        for line in list_tts_voices(config, args.lang, args.backend):
+            print(line)
+        return 0
+
     parser.print_help()
     return 2
 

src/saiki/config.py

@@ -23,14 +23,15 @@ DEFAULT_CONFIG: dict[str, Any] = {
     "audio_output_root": "~/Languages/Anki/anki-audio",
     "word_output_root": "~/Languages/Anki/anki-words",
     "sentence_dir": "~/Languages/Anki",
+    "tts_model_dir": "~/.local/share/saiki/models",
     "note_model": "Basic",
     "fields": {"front": "Front", "back": "Back"},
     "languages": {
         "jp": {
             "name": "japanese",
             "transcript_code": "ja",
-            "tts_code": "ja",
+            "tts_backend": "edge-tts",
-            "tts_tld": "com",
+            "tts_voice": "ja-JP-NanamiNeural",
             "tts_tempo": 1.35,
             "decks": ["日本語"],
             "word_model": "ja_core_news_lg",
@@ -40,8 +41,8 @@ DEFAULT_CONFIG: dict[str, Any] = {
         "es": {
             "name": "spanish",
             "transcript_code": "es",
-            "tts_code": "es",
+            "tts_backend": "edge-tts",
-            "tts_tld": "es",
+            "tts_voice": "es-ES-ElviraNeural",
             "tts_tempo": 1.25,
             "decks": ["Español"],
             "word_model": "es_core_news_sm",
@@ -54,73 +55,108 @@ DEFAULT_CONFIG: dict[str, Any] = {
 
 @dataclass(frozen=True)
 class Config:
+    """Typed convenience wrapper around the merged YAML configuration.
+
+    The underlying ``data`` mapping remains available for simple serialization
+    and tests, while properties and helpers provide normalized paths and common
+    language-specific lookups for the rest of the application.
+    """
+
     data: dict[str, Any]
 
     @property
     def anki_connect_url(self) -> str:
+        """URL for the local AnkiConnect HTTP server."""
         return str(self.data["anki_connect_url"])
 
     @property
     def media_dir(self) -> str:
+        """Expanded path to Anki's collection.media directory."""
         return expand_path(str(self.data["media_dir"]))
 
     @property
     def audio_output_root(self) -> str:
+        """Expanded root directory for exported listening audio."""
         return expand_path(str(self.data["audio_output_root"]))
 
     @property
     def word_output_root(self) -> str:
+        """Expanded root directory for generated vocabulary lists."""
         return expand_path(str(self.data["word_output_root"]))
 
     @property
     def sentence_dir(self) -> str:
+        """Expanded directory used for relative sentence import files."""
         return expand_path(str(self.data["sentence_dir"]))
 
+    @property
+    def tts_model_dir(self) -> str:
+        """Expanded directory used to resolve local TTS model paths."""
+        return expand_path(str(self.data["tts_model_dir"]))
+
     @property
     def note_model(self) -> str:
+        """Anki note type used when importing generated sentence cards."""
         return str(self.data.get("note_model", "Basic"))
 
     @property
     def fields(self) -> dict[str, str]:
+        """Configured logical field names, currently front and back."""
         return dict(self.data.get("fields", {}))
 
     @property
     def languages(self) -> dict[str, dict[str, Any]]:
+        """Language configurations keyed by CLI language code."""
         return dict(self.data.get("languages", {}))
 
     def language(self, lang: str) -> dict[str, Any]:
+        """Return one language config with shared TTS defaults applied.
+
+        A fresh dict is returned so callers may layer CLI overrides onto it
+        without mutating the loaded configuration.
+        """
         try:
-            return dict(self.languages[lang])
+            language = dict(self.languages[lang])
+            language.setdefault("tts_model_dir", self.tts_model_dir)
+            return language
         except KeyError as e:
             available = ", ".join(sorted(self.languages))
             raise ValueError(f"Unsupported language '{lang}'. Available: {available}") from e
 
     def language_name(self, lang: str) -> str:
+        """Return the long language bucket name for output directories."""
         return str(self.language(lang)["name"])
 
     def transcript_code(self, lang: str) -> str:
+        """Return the language code expected by transcript providers."""
         return str(self.language(lang)["transcript_code"])
 
     def decks_for(self, lang: str) -> list[str]:
+        """Return configured Anki deck names for a language."""
         return list(self.language(lang).get("decks", []))
 
     def field_for(self, lang: str) -> str:
+        """Return the Anki field to mine for vocabulary."""
         return str(self.language(lang).get("field", self.fields.get("back", "Back")))
 
     def sentence_file_for(self, lang: str) -> str:
+        """Resolve the sentence import file for a language."""
         value = str(self.language(lang).get("sentence_file", f"sentences_{lang}.txt"))
         return expand_path(value if os.path.isabs(value) or value.startswith("~") else os.path.join(self.sentence_dir, value))
 
 
 def expand_path(path: str) -> str:
+    """Expand ``~`` and environment variables in a configured path."""
     return os.path.expanduser(os.path.expandvars(path))
 
 
 def default_config_path() -> str:
+    """Return the conventional user config path."""
     return expand_path("~/.config/saiki/config.yaml")
 
 
 def deep_merge(base: dict[str, Any], override: dict[str, Any]) -> dict[str, Any]:
+    """Recursively merge a user config mapping over default config values."""
     result = copy.deepcopy(base)
     for key, value in override.items():
         if isinstance(value, dict) and isinstance(result.get(key), dict):
@@ -131,6 +167,7 @@ def deep_merge(base: dict[str, Any], override: dict[str, Any]) -> dict[str, Any]
 
 
 def load_config(path: str | None = None) -> Config:
+    """Load defaults plus an optional YAML config file."""
     config = copy.deepcopy(DEFAULT_CONFIG)
     config_path = expand_path(path) if path else default_config_path()
     if os.path.exists(config_path):
@@ -145,4 +182,5 @@ def load_config(path: str | None = None) -> Config:
 
 
 def language_choices(config: Config) -> list[str]:
+    """Return sorted language codes suitable for argparse choices."""
     return sorted(config.languages.keys())

src/saiki/importer.py

@@ -1,28 +1,57 @@
-"""Generate TTS audio and add sentence notes to Anki."""
+"""Generate TTS audio and add sentence notes to Anki.
+
+This module owns the TTS backend abstraction used by both ``import`` and
+``tts-test``. Backends synthesize their native output format first, then ffmpeg
+normalizes the result to MP3 and applies the configured tempo multiplier.
+"""
 
 from __future__ import annotations
 
-import os
 import csv
+import os
 import shutil
 import subprocess
 import tempfile
 import time
-from dataclasses import dataclass
+from dataclasses import dataclass, field
-from typing import Callable
+from typing import Any, Callable, Mapping
 
 from .ankiconnect import anki_request
-from .config import Config
+from .config import Config, expand_path
 
 
 @dataclass(frozen=True)
 class ImportResult:
+    """Summary of one sentence import run."""
+
     processed: int
     added: int
     failed: int
+    errors: list[str] = field(default_factory=list)
+
+
+@dataclass(frozen=True)
+class PreparedTtsBackend:
+    """Runtime-ready TTS backend callable plus its native audio extension."""
+
+    name: str
+    raw_ext: str
+    synthesize: Callable[[str, str], None]
+
+
+@dataclass(frozen=True)
+class TtsBackendSpec:
+    """Static metadata needed to validate and build a TTS backend."""
+
+    raw_ext: str
+    build: Callable[[dict[str, Any]], Callable[[str, str], None]]
+    required_keys: tuple[str, ...] = ()
+    command: str | None = None
+    list_voices: Callable[[dict[str, Any]], list[str]] | None = None
 
 
 def parse_tags(value: str | None) -> list[str]:
+    """Parse comma-separated tag text and add Saiki's default TTS tags."""
     tags = ["text-to-speech"]
     if value:
         tags.extend(tag.strip() for tag in value.split(",") if tag.strip())
@@ -32,15 +61,407 @@ def parse_tags(value: str | None) -> list[str]:
 
 
 def require_command(name: str) -> None:
+    """Raise a friendly error if an external command is not on PATH."""
     if shutil.which(name) is None:
         raise RuntimeError(f"Required command not found: {name}")
 
 
-def generate_tts(sentence: str, raw_output: str, lang_code: str, tld: str) -> None:
+_TTS_PATH_KEYS = ("tts_model", "tts_voices", "tts_vocab_config", "tts_config")
-    subprocess.run(["gtts-cli", sentence, "--lang", lang_code, "--tld", tld, "--output", raw_output], check=True)
+_MAX_ERROR_DETAILS = 5
+_DEFAULT_TEST_TEXT = {
+    "jp": "これはテストです。",
+    "es": "Esta es una prueba.",
+}
+
+
+def _generate_gtts(cfg: dict[str, Any]) -> Callable[[str, str], None]:
+    """Build a gTTS synthesizer using the command-line wrapper."""
+    lang_code = str(cfg["tts_code"])
+    tld = str(cfg["tts_tld"])
+
+    def synthesize(sentence: str, output: str) -> None:
+        subprocess.run(
+            ["gtts-cli", sentence, "--lang", lang_code, "--tld", tld, "--output", output],
+            stdin=subprocess.DEVNULL,
+            check=True,
+        )
+
+    return synthesize
+
+
+def _generate_edge_tts(cfg: dict[str, Any]) -> Callable[[str, str], None]:
+    """Build an edge-tts synthesizer for a configured neural voice."""
+    voice = str(cfg["tts_voice"])
+
+    def synthesize(sentence: str, output: str) -> None:
+        subprocess.run(
+            ["edge-tts", "--voice", voice, "--text", sentence, "--write-media", output],
+            stdin=subprocess.DEVNULL,
+            check=True,
+        )
+
+    return synthesize
+
+
+def _generate_piper(cfg: dict[str, Any]) -> Callable[[str, str], None]:
+    """Build a Piper synthesizer around a local ONNX voice model."""
+    model = str(cfg["tts_model"])
+    _require_file("piper", "tts_model", model)
+    config = str(cfg["tts_config"]) if cfg.get("tts_config") else None
+    if config:
+        _require_file("piper", "tts_config", config)
+
+    def synthesize(sentence: str, output: str) -> None:
+        command = ["piper", "--model", model]
+        if config:
+            command.extend(["--config", config])
+        command.extend(["--output_file", output])
+        subprocess.run(
+            command,
+            input=f"{sentence}\n".encode("utf-8"),
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            check=True,
+        )
+
+    return synthesize
+
+
+def _generate_espeak(cfg: dict[str, Any]) -> Callable[[str, str], None]:
+    """Build an espeak-ng synthesizer for a configured voice code."""
+    voice = str(cfg["tts_voice"])
+
+    def synthesize(sentence: str, output: str) -> None:
+        subprocess.run(["espeak-ng", "-v", voice, "-w", output, sentence], stdin=subprocess.DEVNULL, check=True)
+
+    return synthesize
+
+
+def _generate_kokoro(cfg: dict[str, Any]) -> Callable[[str, str], None]:
+    """Build a Kokoro ONNX synthesizer from local model and voice files."""
+    _require_file("kokoro", "tts_model", str(cfg["tts_model"]))
+    _require_file("kokoro", "tts_voices", str(cfg["tts_voices"]))
+    if cfg.get("tts_vocab_config"):
+        _require_file("kokoro", "tts_vocab_config", str(cfg["tts_vocab_config"]))
+
+    try:
+        from kokoro_onnx import Kokoro  # type: ignore
+        import soundfile as sf  # type: ignore
+    except ImportError as exc:
+        raise RuntimeError(
+            "kokoro backend requires 'kokoro-onnx' and 'soundfile'. Install them first."
+        ) from exc
+
+    kokoro_kwargs = {}
+    if cfg.get("tts_vocab_config"):
+        kokoro_kwargs["vocab_config"] = str(cfg["tts_vocab_config"])
+    kokoro = Kokoro(str(cfg["tts_model"]), str(cfg["tts_voices"]), **kokoro_kwargs)
+    voice = str(cfg["tts_voice"])
+    lang_code = str(cfg["tts_code"])
+    speed = _optional_float(cfg, "tts_speed", 1.0)
+
+    def synthesize(sentence: str, output: str) -> None:
+        samples, sample_rate = kokoro.create(sentence, voice=voice, speed=speed, lang=lang_code)
+        sf.write(output, samples, sample_rate)
+
+    return synthesize
+
+
+def _list_gtts_voices(cfg: dict[str, Any]) -> list[str]:
+    """Return voice-listing guidance for gTTS."""
+    return [
+        "gtts does not expose named voices.",
+        f"Current settings: tts_code={cfg.get('tts_code', '<unset>')}, tts_tld={cfg.get('tts_tld', '<unset>')}",
+    ]
+
+
+def _list_edge_voices(cfg: dict[str, Any]) -> list[str]:
+    """Return the configured edge-tts voice or ask the CLI to list online voices."""
+    if cfg.get("tts_voice"):
+        return [
+            f"Configured edge-tts voice: {cfg['tts_voice']}",
+            "Run `edge-tts --list-voices` directly to browse the full online voice catalog.",
+        ]
+    return _run_voice_command(["edge-tts", "--list-voices"])
+
+
+def _list_espeak_voices(cfg: dict[str, Any]) -> list[str]:
+    """List espeak-ng voices, narrowed by the configured language when possible."""
+    voice_filter = str(cfg.get("tts_voice") or cfg.get("tts_code") or "")
+    arg = f"--voices={voice_filter}" if voice_filter else "--voices"
+    return _run_voice_command(["espeak-ng", arg])
+
+
+def _list_piper_voices(cfg: dict[str, Any]) -> list[str]:
+    """Return Piper model guidance instead of pretending it has a voice catalog."""
+    model = cfg.get("tts_model")
+    if model:
+        return [f"Configured Piper model: {model}"]
+    return ["Piper voices are model files. Set tts_model to a downloaded .onnx voice model."]
+
+
+def _list_kokoro_voices(cfg: dict[str, Any]) -> list[str]:
+    """Return Kokoro voice-bundle guidance from the configured files."""
+    voice = cfg.get("tts_voice")
+    voices = cfg.get("tts_voices")
+    if voice or voices:
+        return [f"Configured Kokoro voice: {voice or '<unset>'}", f"Voice bundle: {voices or '<unset>'}"]
+    return ["Kokoro voices come from the configured tts_voices bundle. Set tts_voice to one voice from it."]
+
+
+# Registry entries describe validation, dependency checks, synthesis, and
+# optional voice-listing behavior in one place so new free backends can be added
+# without changing the CLI or import workflows.
+_TTS_BACKENDS: dict[str, TtsBackendSpec] = {
+    "gtts": TtsBackendSpec(
+        raw_ext=".mp3",
+        command="gtts-cli",
+        required_keys=("tts_code", "tts_tld"),
+        build=_generate_gtts,
+        list_voices=_list_gtts_voices,
+    ),
+    "edge-tts": TtsBackendSpec(
+        raw_ext=".mp3",
+        command="edge-tts",
+        required_keys=("tts_voice",),
+        build=_generate_edge_tts,
+        list_voices=_list_edge_voices,
+    ),
+    "piper": TtsBackendSpec(
+        raw_ext=".wav",
+        command="piper",
+        required_keys=("tts_model",),
+        build=_generate_piper,
+        list_voices=_list_piper_voices,
+    ),
+    "espeak-ng": TtsBackendSpec(
+        raw_ext=".wav",
+        command="espeak-ng",
+        required_keys=("tts_voice",),
+        build=_generate_espeak,
+        list_voices=_list_espeak_voices,
+    ),
+    "kokoro": TtsBackendSpec(
+        raw_ext=".wav",
+        required_keys=("tts_model", "tts_voices", "tts_voice", "tts_code"),
+        build=_generate_kokoro,
+        list_voices=_list_kokoro_voices,
+    ),
+}
+
+
+def supported_tts_backends() -> list[str]:
+    """Return supported backend names for argparse choices and error messages."""
+    return sorted(_TTS_BACKENDS)
+
+
+def prepare_tts_backend(lang_cfg: dict[str, Any]) -> PreparedTtsBackend:
+    """Validate config and return a callable backend for one language.
+
+    Path-like config values are expanded before validation, required keys are
+    checked per backend, and external command dependencies are verified when a
+    backend shells out to a CLI tool.
+    """
+    backend = str(lang_cfg.get("tts_backend", "gtts")).strip()
+    spec = _TTS_BACKENDS.get(backend)
+    if spec is None:
+        raise ValueError(_unknown_backend_message(backend))
+
+    cfg = _expand_tts_paths(lang_cfg)
+    _require_backend_keys(backend, cfg, spec.required_keys)
+    if spec.command:
+        require_command(spec.command)
+    return PreparedTtsBackend(name=backend, raw_ext=spec.raw_ext, synthesize=spec.build(cfg))
+
+
+def list_tts_voices(config: Config, lang: str | None = None, backend: str | None = None) -> list[str]:
+    """Return voice names or backend-specific hints for the selected TTS backend."""
+    lang_cfg = config.language(lang) if lang else {}
+    if backend:
+        lang_cfg = {**lang_cfg, "tts_backend": backend}
+    name = str(lang_cfg.get("tts_backend", "gtts")).strip()
+    spec = _TTS_BACKENDS.get(name)
+    if spec is None:
+        raise ValueError(_unknown_backend_message(name))
+    cfg = _expand_tts_paths(lang_cfg)
+    if spec.list_voices is None:
+        return [f"{name} does not support voice listing."]
+    return spec.list_voices(cfg)
+
+
+def default_tts_test_text(lang: str) -> str:
+    """Return a short built-in phrase for ``tts-test``."""
+    return _DEFAULT_TEST_TEXT.get(lang, "This is a test.")
+
+
+def synthesize_tts_sample(
+    config: Config,
+    lang: str,
+    text: str | None = None,
+    output: str | None = None,
+    tts_overrides: Mapping[str, Any] | None = None,
+) -> str:
+    """Generate one TTS sample without touching Anki.
+
+    This is the safest way to verify backend configuration. It uses the same
+    backend preparation and ffmpeg normalization path as real imports.
+    """
+    language = _language_config(config, lang, tts_overrides)
+    backend = prepare_tts_backend(language)
+    tempo = _tts_tempo(language)
+    require_command("ffmpeg")
+
+    output_path = expand_path(output) if output else _default_tts_output(lang, backend.name)
+    output_dir = os.path.dirname(output_path)
+    if output_dir:
+        os.makedirs(output_dir, exist_ok=True)
+
+    sentence = text or default_tts_test_text(lang)
+    with tempfile.TemporaryDirectory() as temp_dir:
+        raw_output = os.path.join(temp_dir, f"tts_test_original{backend.raw_ext}")
+        backend.synthesize(sentence, raw_output)
+        speed_audio(raw_output, output_path, tempo)
+    return output_path
+
+
+def _raw_ext(backend: str) -> str:
+    """Return a backend's raw extension, defaulting to MP3 for unknown names."""
+    spec = _TTS_BACKENDS.get(backend)
+    return spec.raw_ext if spec else ".mp3"
+
+
+def _language_config(
+    config: Config,
+    lang: str,
+    tts_overrides: Mapping[str, Any] | None = None,
+) -> dict[str, Any]:
+    """Return a language config with any non-None CLI overrides applied."""
+    language = config.language(lang)
+    if tts_overrides:
+        language.update({key: value for key, value in tts_overrides.items() if value is not None})
+    return language
+
+
+def _default_tts_output(lang: str, backend: str) -> str:
+    """Return the default sample output path for ``tts-test``."""
+    safe_backend = backend.replace(os.sep, "_")
+    return os.path.abspath(f"tts_test_{lang}_{safe_backend}.mp3")
+
+
+def _expand_tts_paths(lang_cfg: Mapping[str, Any]) -> dict[str, Any]:
+    """Expand TTS paths and resolve relative model files under tts_model_dir."""
+    cfg = dict(lang_cfg)
+    if isinstance(cfg.get("tts_model_dir"), str):
+        cfg["tts_model_dir"] = expand_path(str(cfg["tts_model_dir"]))
+    for key in _TTS_PATH_KEYS:
+        if isinstance(cfg.get(key), str):
+            path = expand_path(str(cfg[key]))
+            if not os.path.isabs(path) and cfg.get("tts_model_dir"):
+                path = os.path.join(str(cfg["tts_model_dir"]), path)
+            cfg[key] = path
+    return cfg
+
+
+def _require_backend_keys(backend: str, cfg: Mapping[str, Any], keys: tuple[str, ...]) -> None:
+    """Ensure backend-specific required config keys are present and non-empty."""
+    missing = [key for key in keys if cfg.get(key) is None or (isinstance(cfg.get(key), str) and not cfg[key].strip())]
+    if missing:
+        raise RuntimeError(f"{backend} backend requires config key(s): {', '.join(missing)}")
+
+
+def _require_file(backend: str, key: str, path: str) -> None:
+    """Ensure a configured model path exists before calling a backend."""
+    if not os.path.isfile(path):
+        raise RuntimeError(f"{backend} backend config key {key} points to a missing file: {path}")
+
+
+def format_tts_error(exc: Exception) -> str:
+    """Format backend and ffmpeg failures for concise CLI output."""
+    return _error_message(exc)
+
+
+def _unknown_backend_message(backend: str) -> str:
+    """Build a consistent unknown-backend error message."""
+    return f"Unknown TTS backend: {backend!r}. Choose from: {', '.join(supported_tts_backends())}"
+
+
+def _optional_float(cfg: Mapping[str, Any], key: str, default: float | None) -> float | None:
+    """Parse an optional numeric config value."""
+    if cfg.get(key) is None:
+        return default
+    try:
+        return float(cfg[key])
+    except (TypeError, ValueError) as exc:
+        raise RuntimeError(f"{key} must be a number.") from exc
+
+
+def _run_voice_command(command: list[str], timeout: float = 30.0) -> list[str]:
+    """Run an external voice-listing command and return printable lines."""
+    if shutil.which(command[0]) is None:
+        return [f"Required command not found: {command[0]}"]
+    try:
+        result = subprocess.run(
+            command,
+            text=True,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            timeout=timeout,
+            check=True,
+        )
+    except subprocess.CalledProcessError as exc:
+        return [f"{' '.join(command)} failed: {_subprocess_detail(exc)}"]
+    except subprocess.TimeoutExpired:
+        return [f"{' '.join(command)} timed out after {timeout:g}s."]
+    lines = [line for line in result.stdout.splitlines() if line.strip()]
+    return lines or ["No voices returned."]
+
+
+def _short_text(value: str, limit: int = 500) -> str:
+    """Collapse and truncate long subprocess output for display."""
+    text = " ".join(value.strip().split())
+    if len(text) <= limit:
+        return text
+    return f"{text[:limit - 3]}..."
+
+
+def _sentence_label(sentence: str, limit: int = 80) -> str:
+    """Return a compact representation of a sentence for error lists."""
+    text = " ".join(sentence.split())
+    if len(text) > limit:
+        text = f"{text[:limit - 3]}..."
+    return repr(text)
+
+
+def _subprocess_detail(exc: subprocess.CalledProcessError) -> str:
+    """Extract useful stdout/stderr context from a failed subprocess."""
+    stderr = exc.stderr
+    stdout = exc.stdout
+    detail = stderr if stderr else stdout
+    if isinstance(detail, bytes):
+        detail = detail.decode("utf-8", errors="replace")
+    text = _short_text(str(detail or ""))
+    command = exc.cmd if isinstance(exc.cmd, str) else " ".join(str(part) for part in exc.cmd)
+    suffix = f": {text}" if text else ""
+    return f"{command} exited with status {exc.returncode}{suffix}"
+
+
+def _error_message(exc: Exception) -> str:
+    """Convert an exception into a short user-facing string."""
+    if isinstance(exc, subprocess.CalledProcessError):
+        return _subprocess_detail(exc)
+    return _short_text(str(exc) or exc.__class__.__name__)
+
+
+def _tts_tempo(cfg: Mapping[str, Any]) -> float:
+    """Validate and return the post-processing tempo multiplier."""
+    tempo = _optional_float(cfg, "tts_tempo", 1.0)
+    if tempo is None or tempo <= 0:
+        raise RuntimeError("tts_tempo must be greater than 0.")
+    return tempo
 
 
 def speed_audio(raw_output: str, output_path: str, tempo: float) -> None:
+    """Convert backend output to an MP3 and apply ffmpeg's atempo filter."""
     subprocess.run(
         ["ffmpeg", "-loglevel", "error", "-i", raw_output, "-filter:a", f"atempo={tempo}", "-y", output_path],
         stdin=subprocess.DEVNULL,
@@ -49,6 +470,11 @@ def speed_audio(raw_output: str, output_path: str, tempo: float) -> None:
 
 
 def read_sentences(path: str) -> list[str]:
+    """Read sentences from plain text, CSV, or TSV input.
+
+    CSV and TSV imports must contain a ``sentence`` header so exports from the
+    YouTube sentence-mining command can be imported directly.
+    """
     expanded = os.path.expanduser(path)
     if expanded.lower().endswith((".tsv", ".csv")):
         delimiter = "\t" if expanded.lower().endswith(".tsv") else ","
@@ -68,11 +494,19 @@ def import_sentences(
     sentence_file: str | None = None,
     tags_value: str | None = None,
     request: Callable = anki_request,
+    tts_overrides: Mapping[str, Any] | None = None,
 ) -> ImportResult:
-    require_command("gtts-cli")
+    """Generate TTS for each sentence and add cards through AnkiConnect.
+
+    The first configured deck for the language is used as the destination.
+    Audio is attached to the front field so Anki imports the temporary MP3 into
+    its media collection before the temporary directory is removed.
+    """
+    language = _language_config(config, lang, tts_overrides)
+    backend = prepare_tts_backend(language)
+    tempo = _tts_tempo(language)
     require_command("ffmpeg")
 
-    language = config.language(lang)
     decks = list(language.get("decks", []))
     if not decks:
         raise RuntimeError(f"No deck configured for language: {lang}")
@@ -85,15 +519,16 @@ def import_sentences(
     back_field = config.fields.get("back", "Back")
     added = 0
     failed = 0
+    errors: list[str] = []
 
     with tempfile.TemporaryDirectory() as temp_dir:
         for sentence in sentences:
             basename = f"tts_{time.strftime('%Y%m%d_%H%M%S')}_{lang}_{os.getpid()}_{added + failed}"
-            raw_output = os.path.join(temp_dir, f"{basename}_original.mp3")
+            raw_output = os.path.join(temp_dir, f"{basename}_original{backend.raw_ext}")
             output_path = os.path.join(temp_dir, f"{basename}.mp3")
             try:
-                generate_tts(sentence, raw_output, str(language["tts_code"]), str(language["tts_tld"]))
+                backend.synthesize(sentence, raw_output)
-                speed_audio(raw_output, output_path, float(language["tts_tempo"]))
+                speed_audio(raw_output, output_path, tempo)
                 request(
                     "addNote",
                     url=config.anki_connect_url,
@@ -107,6 +542,8 @@ def import_sentences(
                     },
                 )
                 added += 1
-            except Exception:
+            except Exception as exc:
                 failed += 1
-    return ImportResult(processed=len(sentences), added=added, failed=failed)
+                if len(errors) < _MAX_ERROR_DETAILS:
+                    errors.append(f"{_sentence_label(sentence)}: {_error_message(exc)}")
+    return ImportResult(processed=len(sentences), added=added, failed=failed, errors=errors)

src/saiki/text.py

@@ -8,6 +8,11 @@ import regex as re
 
 
 def extract_first_visible_line(text: str) -> str:
+    """Strip simple HTML and return the first visible line of text.
+
+    This is useful for card fields where the first line is the sentence and
+    subsequent lines may contain notes, hints, or generated markup.
+    """
     text = unescape(text or "")
     text = re.sub(r"</?(br|div|p)[^>]*>", "\n", text, flags=re.IGNORECASE)
     text = re.sub(r"<[^>]+>", "", text)
@@ -16,6 +21,7 @@ def extract_first_visible_line(text: str) -> str:
 
 
 def extract_visible_text(text: str) -> str:
+    """Strip simple HTML while preserving paragraph-like line breaks."""
     text = unescape(text or "")
     text = re.sub(r"</?(br|div|p)[^>]*>", "\n", text, flags=re.IGNORECASE)
     text = re.sub(r"<[^>]+>", "", text)
@@ -25,5 +31,5 @@ def extract_visible_text(text: str) -> str:
 
 
 def normalize_word_key(value: str) -> str:
+    """Normalize a word for set membership and comparison."""
     return re.sub(r"\s+", " ", value.strip().lower())
-

src/saiki/words.py

@@ -26,15 +26,22 @@ JAPANESE_ALLOWED_POS = {"NOUN", "PROPN", "VERB", "ADJ"}
 
 
 def setup_logging(logfile: str) -> None:
+    """Configure file logging for word extraction scripts."""
     os.makedirs(os.path.dirname(os.path.abspath(logfile)), exist_ok=True)
     logging.basicConfig(filename=logfile, level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s")
 
 
 def build_query_from_decks(decks: list[str]) -> str:
+    """Build an Anki search query that matches any configured deck."""
     return " OR ".join(f'deck:"{d}"' for d in decks)
 
 
 def japanese_filter(token) -> bool:
+    """Return whether a spaCy token is useful Japanese vocabulary.
+
+    The filter is intentionally conservative: it keeps content words and drops
+    common particles, helper grammar, stop words, URLs, and obvious HTML debris.
+    """
     text = (token.text or "").strip()
     lemma = (token.lemma_ or "").strip()
     if not text or not JAPANESE_CHAR_RE.fullmatch(text):
@@ -51,14 +58,17 @@ def japanese_filter(token) -> bool:
 
 
 def spanish_filter(token) -> bool:
+    """Return whether a spaCy token is useful Spanish vocabulary."""
     return bool(getattr(token, "is_alpha", False)) and not bool(getattr(token, "is_stop", False))
 
 
 def spanish_format(token) -> str:
+    """Normalize a Spanish token to its lowercase lemma."""
     return (token.lemma_ or token.text or "").lower().strip()
 
 
 def japanese_format(token) -> str:
+    """Format a Japanese token as lemma plus surface form when they differ."""
     lemma = (token.lemma_ or "").strip()
     surface = (token.text or "").strip()
     if lemma and surface and lemma != surface:
@@ -73,6 +83,7 @@ LANGUAGE_PROFILES = {
 
 
 def load_spacy_model(model_name: str):
+    """Load a spaCy model with installation-oriented error messages."""
     try:
         import spacy  # type: ignore
     except Exception as e:
@@ -84,6 +95,7 @@ def load_spacy_model(model_name: str):
 
 
 def get_notes(query: str, config: Config, request: Callable = anki_request) -> list[dict]:
+    """Fetch Anki note details matching a search query."""
     note_ids = request("findNotes", url=config.anki_connect_url, query=query) or []
     if not note_ids:
         return []
@@ -98,6 +110,7 @@ def extract_counts(
     output_format: Callable,
     use_full_field: bool,
 ) -> Counter:
+    """Count formatted vocabulary items across Anki notes."""
     counter: Counter = Counter()
     for note in notes:
         fields = note.get("fields", {}) or {}
@@ -114,6 +127,7 @@ def extract_counts(
 
 
 def write_counts(counter: Counter, out_path: str, min_freq: int) -> int:
+    """Write a sorted ``word frequency`` list and return the number of rows."""
     items = [(w, c) for (w, c) in counter.items() if c >= min_freq]
     items.sort(key=lambda x: (-x[1], x[0]))
     os.makedirs(os.path.dirname(os.path.abspath(out_path)), exist_ok=True)
@@ -124,6 +138,7 @@ def write_counts(counter: Counter, out_path: str, min_freq: int) -> int:
 
 
 def read_word_file(path: str) -> set[str]:
+    """Read a ``word frequency`` file into normalized word keys."""
     words: set[str] = set()
     with open(os.path.expanduser(path), "r", encoding="utf-8") as fh:
         for line in fh:
@@ -136,6 +151,7 @@ def read_word_file(path: str) -> set[str]:
 
 
 def compare_word_files(source_path: str, known_path: str) -> list[str]:
+    """Return source rows whose normalized word is not in the known list."""
     known = read_word_file(known_path)
     new_words: list[str] = []
     with open(os.path.expanduser(source_path), "r", encoding="utf-8") as fh:
@@ -162,6 +178,12 @@ def extract_words(
     spacy_model: str | None = None,
     request: Callable = anki_request,
 ) -> dict[str, object]:
+    """Extract frequent vocabulary from configured Anki cards.
+
+    The function accepts explicit query/deck/field overrides for CLI use, but
+    defaults to the selected language config. Its dictionary return value keeps
+    the CLI output simple and gives tests stable fields to assert against.
+    """
     language_bucket = config.language_name(lang)
     profile = LANGUAGE_PROFILES[language_bucket]
     search_query = query or build_query_from_decks(decks or config.decks_for(lang))
@@ -180,4 +202,3 @@ def extract_words(
     counter = extract_counts(notes, field_name, nlp, profile["token_filter"], profile["output_format"], full_field)
     written = write_counts(counter, out_path, min_freq)
     return {"query": search_query, "notes": len(notes), "unique": len(counter), "written": written, "out": out_path}
-

src/saiki/youtube.py

@@ -27,11 +27,14 @@ STOPWORDS = {
 
 @dataclass(frozen=True)
 class TranscriptLine:
+    """One cleaned transcript line with its start timestamp in seconds."""
+
     start: float
     text: str
 
 
 def extract_video_id(url_or_id: str) -> str:
+    """Extract a YouTube video id from a URL, or pass an id through unchanged."""
     if "youtube" in url_or_id or "youtu.be" in url_or_id:
         query = urlparse(url_or_id)
         if query.hostname == "youtu.be":
@@ -44,11 +47,13 @@ def extract_video_id(url_or_id: str) -> str:
 
 
 def video_url(video_or_id: str) -> str:
+    """Return a canonical watch URL for a YouTube id or URL."""
     video_id = extract_video_id(video_or_id)
     return f"https://www.youtube.com/watch?v={video_id}"
 
 
 def fetch_transcript(video_id: str, lang_code: str):
+    """Fetch a transcript while supporting old and new library APIs."""
     if hasattr(YouTubeTranscriptApi, "fetch"):
         api = YouTubeTranscriptApi()
         return api.fetch(video_id, languages=[lang_code])
@@ -58,18 +63,21 @@ def fetch_transcript(video_id: str, lang_code: str):
 
 
 def snippet_text(entry) -> str:
+    """Read transcript text from either dict-like or object-like entries."""
     if isinstance(entry, dict):
         return entry.get("text", "") or ""
     return getattr(entry, "text", "") or ""
 
 
 def snippet_start(entry) -> float:
+    """Read transcript start time from either dict-like or object-like entries."""
     if isinstance(entry, dict):
         return float(entry.get("start", 0.0) or 0.0)
     return float(getattr(entry, "start", 0.0) or 0.0)
 
 
 def transcript_lines(entries) -> list[TranscriptLine]:
+    """Normalize raw transcript entries into non-empty transcript lines."""
     lines: list[TranscriptLine] = []
     for entry in entries:
         text = snippet_text(entry).replace("\n", " ").strip()
@@ -79,6 +87,7 @@ def transcript_lines(entries) -> list[TranscriptLine]:
 
 
 def tokenize_japanese(text: str) -> list[str]:
+    """Tokenize Japanese text with fugashi."""
     try:
         from fugashi import Tagger
     except ImportError as e:
@@ -88,15 +97,18 @@ def tokenize_japanese(text: str) -> list[str]:
 
 
 def tokenize_spanish(text: str, raw: bool = False) -> list[str]:
+    """Tokenize Spanish-ish text with a lightweight word regex."""
     tokens = re.findall(r"\b[\wáéíóúñü]+\b", text)
     return tokens if raw else [t.lower() for t in tokens]
 
 
 def tokenize_text(text: str, lang_code: str, raw: bool = False) -> list[str]:
+    """Dispatch transcript tokenization by language code."""
     return tokenize_japanese(text) if lang_code == "ja" else tokenize_spanish(text, raw=raw)
 
 
 def count_words(tokens: list[str], lang_code: str, remove_stopwords: bool = True) -> Counter:
+    """Count tokens, optionally excluding the built-in stopword list."""
     if remove_stopwords:
         stopwords = STOPWORDS.get(lang_code, set())
         tokens = [t for t in tokens if t not in stopwords]
@@ -104,6 +116,7 @@ def count_words(tokens: list[str], lang_code: str, remove_stopwords: bool = True
 
 
 def sentence_vocab(sentence: str, lang_code: str, known_words: set[str] | None = None) -> list[str]:
+    """Guess distinct useful vocabulary for one transcript sentence."""
     words: list[str] = []
     seen: set[str] = set()
     for token in tokenize_text(sentence, lang_code):
@@ -126,6 +139,7 @@ def write_sentence_export(
     known_words_path: str | None = None,
     only_new: bool = False,
 ) -> int:
+    """Write transcript lines as Anki-importable sentence rows."""
     known = read_word_file(known_words_path) if known_words_path else None
     os.makedirs(os.path.dirname(os.path.abspath(out_path)), exist_ok=True)
     written = 0
@@ -154,6 +168,7 @@ def run_youtube(
     known_words: str | None = None,
     only_new: bool = False,
 ) -> dict[str, object]:
+    """Run transcript mining in either vocabulary or sentence-export mode."""
     lang_code = config.transcript_code(lang)
     video_id = extract_video_id(video)
     entries = fetch_transcript(video_id, lang_code)
@@ -176,4 +191,3 @@ def run_youtube(
             for word, count in items:
                 fh.write(f"{word} {count}\n")
     return {"mode": mode, "items": items, "out": out}
-

tests/test_core.py

@@ -4,15 +4,26 @@ import os
 import sys
 import tempfile
 import unittest
+from copy import deepcopy
 from pathlib import Path
+from unittest.mock import patch
 
 SRC_DIR = Path(__file__).resolve().parents[1] / "src"
 if str(SRC_DIR) not in sys.path:
     sys.path.insert(0, str(SRC_DIR))
 
 from saiki.audio import build_playlist, resolve_media_paths
-from saiki.config import DEFAULT_CONFIG, deep_merge
+from saiki.config import Config, DEFAULT_CONFIG, deep_merge
-from saiki.importer import parse_tags, read_sentences
+from saiki.importer import (
+    PreparedTtsBackend,
+    import_sentences,
+    list_tts_voices,
+    parse_tags,
+    prepare_tts_backend,
+    read_sentences,
+    synthesize_tts_sample,
+    supported_tts_backends,
+)
 from saiki.text import extract_first_visible_line, extract_visible_text
 from saiki.words import build_query_from_decks, compare_word_files, read_word_file
 from saiki.youtube import TranscriptLine, extract_video_id, sentence_vocab, write_sentence_export
@@ -23,6 +34,7 @@ class ConfigTests(unittest.TestCase):
         merged = deep_merge(DEFAULT_CONFIG, {"languages": {"es": {"decks": ["Spanish"]}}})
         self.assertEqual(merged["languages"]["es"]["decks"], ["Spanish"])
         self.assertEqual(merged["languages"]["es"]["transcript_code"], "es")
+        self.assertEqual(merged["tts_model_dir"], "~/.local/share/saiki/models")
         self.assertIn("jp", merged["languages"])
 
 
@@ -107,6 +119,87 @@ class ImporterTests(unittest.TestCase):
                 fh.write("Hola mundo\t1.00\thttps://example.test\tmundo\n")
             self.assertEqual(read_sentences(path), ["Hola mundo"])
 
+    def test_supported_tts_backends_are_free_options(self):
+        self.assertEqual(supported_tts_backends(), ["edge-tts", "espeak-ng", "gtts", "kokoro", "piper"])
+
+    def test_list_gtts_voice_hint(self):
+        config = Config(deep_merge(deepcopy(DEFAULT_CONFIG), {"languages": {"es": {"tts_code": "es", "tts_tld": "es"}}}))
+        voices = list_tts_voices(config, "es", backend="gtts")
+        self.assertIn("gtts does not expose named voices.", voices)
+        self.assertIn("tts_code=es", voices[1])
+
+    def test_list_edge_voice_uses_configured_voice(self):
+        voices = list_tts_voices(Config(deepcopy(DEFAULT_CONFIG)), "es")
+        self.assertEqual(voices[0], "Configured edge-tts voice: es-ES-ElviraNeural")
+
+    def test_prepare_tts_backend_validates_required_keys(self):
+        with self.assertRaisesRegex(RuntimeError, "tts_voice"):
+            prepare_tts_backend({"tts_backend": "edge-tts"})
+
+    def test_prepare_tts_backend_expands_model_paths(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            model = os.path.join(tmp, "voice.onnx")
+            config = os.path.join(tmp, "voice.onnx.json")
+            for path in [model, config]:
+                with open(path, "w", encoding="utf-8") as fh:
+                    fh.write("{}")
+
+            with patch("saiki.importer.require_command"):
+                backend = prepare_tts_backend(
+                    {
+                        "tts_backend": "piper",
+                        "tts_model_dir": tmp,
+                        "tts_model": "voice.onnx",
+                        "tts_config": "voice.onnx.json",
+                    }
+                )
+
+            with patch("saiki.importer.subprocess.run") as run:
+                backend.synthesize("Hola", "/tmp/out.wav")
+
+        args = run.call_args.args[0]
+        self.assertEqual(args[2], model)
+        self.assertEqual(args[4], config)
+        self.assertEqual(run.call_args.kwargs["input"], b"Hola\n")
+
+    def test_synthesize_tts_sample_uses_backend_and_speed_audio(self):
+        seen: dict[str, str] = {}
+
+        def synthesize(sentence: str, output: str) -> None:
+            seen["sentence"] = sentence
+            seen["raw_output"] = output
+
+        with tempfile.TemporaryDirectory() as tmp:
+            output = os.path.join(tmp, "sample.mp3")
+            with patch("saiki.importer.prepare_tts_backend") as prepare, patch(
+                "saiki.importer.require_command"
+            ), patch("saiki.importer.speed_audio") as speed:
+                prepare.return_value = PreparedTtsBackend("fake", ".wav", synthesize)
+                result = synthesize_tts_sample(Config(deepcopy(DEFAULT_CONFIG)), "es", output=output)
+
+        self.assertEqual(result, output)
+        self.assertEqual(seen["sentence"], "Esta es una prueba.")
+        self.assertTrue(seen["raw_output"].endswith(".wav"))
+        speed.assert_called_once()
+
+    def test_import_sentences_returns_error_details(self):
+        def fail_synthesis(sentence: str, output: str) -> None:
+            raise RuntimeError("tts broke")
+
+        with tempfile.TemporaryDirectory() as tmp:
+            path = os.path.join(tmp, "sentences.txt")
+            with open(path, "w", encoding="utf-8") as fh:
+                fh.write("Hola mundo\n")
+
+            with patch("saiki.importer.prepare_tts_backend") as prepare, patch("saiki.importer.require_command"):
+                prepare.return_value = PreparedTtsBackend("fake", ".mp3", fail_synthesis)
+                result = import_sentences(Config(deepcopy(DEFAULT_CONFIG)), "es", path, request=lambda *a, **k: None)
+
+        self.assertEqual(result.processed, 1)
+        self.assertEqual(result.added, 0)
+        self.assertEqual(result.failed, 1)
+        self.assertEqual(result.errors, ["'Hola mundo': tts broke"])
+
 
 if __name__ == "__main__":
     unittest.main()