stackvox 0.6.0__tar.gz → 0.8.0__tar.gz

{stackvox-0.6.0/stackvox.egg-info → stackvox-0.8.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: stackvox
-Version: 0.6.0
+Version: 0.8.0
 Summary: Offline TTS library using Kokoro-82M
 Author-email: "StackOne Technologies Ltd." <support@stackone.com>
 License-Expression: Apache-2.0
@@ -88,8 +88,23 @@ stackvox welcome                    # multilingual welcome (6 languages)
 stackvox voices                     # list all voice ids
 echo "from a pipe" | stackvox       # piped stdin works for speak/say
 stackvox speak --file message.txt   # read a whole file
+stackvox paths                      # print resolved cache / socket / pid paths
+stackvox config                     # print resolved default voice/speed/lang
 ```
 
+### Speech normalization
+
+`normalize_for_speech` (Markdown → speakable prose: strip formatting, expand units/numbers, shape pauses, apply a pronunciation dictionary) is available from the shell — no Python import required:
+
+```bash
+stackvox normalize --file post.md               # print normalized text (pipe / debug)
+stackvox normalize --tables csv --strip-emoji < response.md
+stackvox speak --normalize --file post.md       # normalize, then synthesize
+stackvox say   --normalize "**Shipped** 1,198.9 MPG"   # normalize, then send to daemon
+```
+
+`--normalize` is off by default on `speak`/`say`, so notification phrases synthesize exactly as before. The tuning flags mirror the library defaults (all on except emoji-strip): `--no-markdown`, `--no-expand-units`, `--no-expand-numbers`, `--no-pauses`, `--tables {drop,csv}`, `--strip-emoji`, `--no-terminal-stops`, `--locale`, and a `--pronunciations FILE` (a JSON `{"written": "spoken"}` map, applied whole-word and case-insensitively).
+
 Bash completion:
 
 ```bash

{stackvox-0.6.0 → stackvox-0.8.0}/README.md

@@ -54,8 +54,23 @@ stackvox welcome                    # multilingual welcome (6 languages)
 stackvox voices                     # list all voice ids
 echo "from a pipe" | stackvox       # piped stdin works for speak/say
 stackvox speak --file message.txt   # read a whole file
+stackvox paths                      # print resolved cache / socket / pid paths
+stackvox config                     # print resolved default voice/speed/lang
 ```
 
+### Speech normalization
+
+`normalize_for_speech` (Markdown → speakable prose: strip formatting, expand units/numbers, shape pauses, apply a pronunciation dictionary) is available from the shell — no Python import required:
+
+```bash
+stackvox normalize --file post.md               # print normalized text (pipe / debug)
+stackvox normalize --tables csv --strip-emoji < response.md
+stackvox speak --normalize --file post.md       # normalize, then synthesize
+stackvox say   --normalize "**Shipped** 1,198.9 MPG"   # normalize, then send to daemon
+```
+
+`--normalize` is off by default on `speak`/`say`, so notification phrases synthesize exactly as before. The tuning flags mirror the library defaults (all on except emoji-strip): `--no-markdown`, `--no-expand-units`, `--no-expand-numbers`, `--no-pauses`, `--tables {drop,csv}`, `--strip-emoji`, `--no-terminal-stops`, `--locale`, and a `--pronunciations FILE` (a JSON `{"written": "spoken"}` map, applied whole-word and case-insensitively).
+
 Bash completion:
 
 ```bash

{stackvox-0.6.0 → stackvox-0.8.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "stackvox"
-version = "0.6.0"
+version = "0.8.0"
 description = "Offline TTS library using Kokoro-82M"
 readme = "README.md"
 requires-python = ">=3.10"

{stackvox-0.6.0 → stackvox-0.8.0}/stackvox/cli.py

@@ -10,8 +10,9 @@ from pathlib import Path
 
 import soundfile as sf
 
-from stackvox import config, daemon, updates
+from stackvox import config, daemon, paths, updates
 from stackvox.engine import Stackvox
+from stackvox.text import normalize_for_speech
 
 
 def _configure_logging() -> None:
@@ -28,8 +29,11 @@ SUBCOMMANDS = {
     "status",
     "say",
     "speak",
+    "normalize",
     "voices",
     "welcome",
+    "paths",
+    "config",
     "completion",
     "install-helper",
 }
@@ -47,12 +51,12 @@ _stackvox_completion() {
     subcommand="${COMP_WORDS[1]:-}"
 
     if [[ ${COMP_CWORD} -eq 1 ]]; then
-        COMPREPLY=( $(compgen -W "speak say serve stop status voices welcome completion install-helper" -- "$cur") )
+        COMPREPLY=( $(compgen -W "speak say normalize serve stop status voices welcome paths config completion install-helper" -- "$cur") )
         return 0
     fi
 
     case "$prev" in
-        --file|--out)
+        --file|--out|--pronunciations)
             COMPREPLY=( $(compgen -f -- "$cur") )
             return 0
             ;;
@@ -68,14 +72,27 @@ _stackvox_completion() {
             COMPREPLY=( $(compgen -W "en-us en-gb fr-fr it hi pt-br es ja zh" -- "$cur") )
             return 0
             ;;
+        --tables)
+            COMPREPLY=( $(compgen -W "drop csv" -- "$cur") )
+            return 0
+            ;;
+        --locale)
+            COMPREPLY=( $(compgen -W "en-GB" -- "$cur") )
+            return 0
+            ;;
     esac
 
+    local norm_flags="--no-markdown --pronunciations --no-expand-units --no-expand-numbers --no-pauses --tables --strip-emoji --no-terminal-stops --locale"
+
     case "$subcommand" in
         speak)
-            COMPREPLY=( $(compgen -W "--voice --speed --lang --file --out --help" -- "$cur") )
+            COMPREPLY=( $(compgen -W "--voice --speed --lang --file --out --normalize $norm_flags --help" -- "$cur") )
             ;;
         say)
-            COMPREPLY=( $(compgen -W "--voice --speed --lang --file --fallback-say --help" -- "$cur") )
+            COMPREPLY=( $(compgen -W "--voice --speed --lang --file --fallback-say --normalize $norm_flags --help" -- "$cur") )
+            ;;
+        normalize)
+            COMPREPLY=( $(compgen -W "--file $norm_flags --help" -- "$cur") )
             ;;
         serve)
             COMPREPLY=( $(compgen -W "--voice --speed --lang --help" -- "$cur") )
@@ -115,6 +132,7 @@ def _build_parser(defaults: config.Defaults | None = None) -> argparse.ArgumentP
     p_speak.add_argument("text", nargs="?")
     p_speak.add_argument("--file", type=Path)
     p_speak.add_argument("--out", type=Path, help="Write wav instead of playing")
+    _add_normalize_args(p_speak, with_switch=True)
 
     p_say = sub.add_parser("say", help="Send text to daemon (fast; fails if daemon not running)")
     _add_voice_args(p_say, defaults)
@@ -123,6 +141,15 @@ def _build_parser(defaults: config.Defaults | None = None) -> argparse.ArgumentP
     p_say.add_argument(
         "--fallback-say", action="store_true", help="Shell out to macOS `say` if daemon unreachable"
     )
+    _add_normalize_args(p_say, with_switch=True)
+
+    p_normalize = sub.add_parser(
+        "normalize",
+        help="Print speech-normalized text (Markdown -> speakable prose); no synthesis",
+    )
+    p_normalize.add_argument("text", nargs="?")
+    p_normalize.add_argument("--file", type=Path)
+    _add_normalize_args(p_normalize, with_switch=False)
 
     p_serve = sub.add_parser("serve", help="Run the daemon in the foreground")
     _add_voice_args(p_serve, defaults)
@@ -131,6 +158,8 @@ def _build_parser(defaults: config.Defaults | None = None) -> argparse.ArgumentP
     sub.add_parser("status", help="Print daemon status")
     sub.add_parser("voices", help="List available voices")
     sub.add_parser("welcome", help="Play a multilingual welcome message")
+    sub.add_parser("paths", help="Print resolved cache / socket / pid file paths")
+    sub.add_parser("config", help="Print resolved default voice/speed/lang and config path")
 
     p_completion = sub.add_parser("completion", help="Print a shell completion script")
     p_completion.add_argument("shell", choices=["bash"], help="Shell to generate completion for")
@@ -155,6 +184,121 @@ def _add_voice_args(parser: argparse.ArgumentParser, defaults: config.Defaults)
     parser.add_argument("--lang", default=defaults.lang)
 
 
+def _add_normalize_args(parser: argparse.ArgumentParser, *, with_switch: bool) -> None:
+    """Flags mapping one-to-one onto ``stackvox.text.normalize_for_speech``.
+
+    Defaults mirror the library: markdown / units / numbers / pauses /
+    terminal-stops on, emoji-strip off, tables dropped, ``en-GB`` locale.
+
+    ``with_switch`` adds the ``--normalize`` master toggle used by ``speak`` and
+    ``say`` (off by default, so notification phrases synth exactly as before).
+    The ``normalize`` subcommand always normalizes, so it omits the switch.
+    """
+    if with_switch:
+        parser.add_argument(
+            "--normalize",
+            action="store_true",
+            help="Normalize text for speech before synthesizing (see `stackvox normalize`)",
+        )
+    parser.add_argument(
+        "--no-markdown", dest="markdown", action="store_false", help="Treat input as plain text, not Markdown"
+    )
+    parser.add_argument(
+        "--pronunciations",
+        type=Path,
+        help="JSON file mapping written -> spoken forms (whole-word, case-insensitive)",
+    )
+    parser.add_argument(
+        "--no-dev-terms",
+        dest="dev_terms",
+        action="store_false",
+        help="Do not spell out dev acronyms espeak mispronounces (CLI, CI, IDE, AWS, URI, IAM, ...)",
+    )
+    parser.add_argument(
+        "--no-expand-units",
+        dest="expand_units",
+        action="store_false",
+        help="Leave £/p/kWh/MPG/kg/km and ÷ × = as written",
+    )
+    parser.add_argument(
+        "--no-expand-numbers",
+        dest="expand_numbers",
+        action="store_false",
+        help="Leave thousands separators and decimals as written",
+    )
+    parser.add_argument(
+        "--no-pauses",
+        dest="pauses",
+        action="store_false",
+        help="Do not reshape dashes / parentheses into pauses",
+    )
+    parser.add_argument(
+        "--tables",
+        choices=["drop", "csv"],
+        default="drop",
+        help="How Markdown tables are voiced (default: drop)",
+    )
+    parser.add_argument(
+        "--strip-emoji", dest="strip_emoji", action="store_true", help="Remove emoji before speaking"
+    )
+    parser.add_argument(
+        "--no-terminal-stops",
+        dest="terminal_stops",
+        action="store_false",
+        help="Do not add terminal punctuation so a pause lands between lines",
+    )
+    parser.add_argument("--locale", default="en-GB", help="Locale for unit/currency rules (default: en-GB)")
+
+
+def _load_pronunciations(path: Path | None) -> dict[str, str] | None:
+    """Load a ``{written: spoken}`` map from a JSON file.
+
+    Returns ``None`` when no path is given. Raises ``ValueError`` if the file
+    is not a flat table of string -> string entries, so a mistyped dict fails
+    loudly rather than silently voicing the wrong thing.
+    """
+    if path is None:
+        return None
+    import json
+
+    data = json.loads(path.expanduser().read_text(encoding="utf-8"))
+    if not isinstance(data, dict) or not all(
+        isinstance(key, str) and isinstance(value, str) for key, value in data.items()
+    ):
+        raise ValueError(f"{path}: expected a flat JSON object of string -> string entries")
+    return data
+
+
+def _normalize_kwargs(args: argparse.Namespace) -> dict:
+    """Collect the normalize_for_speech keyword arguments from parsed flags."""
+    return {
+        "markdown": args.markdown,
+        "pronunciations": _load_pronunciations(args.pronunciations),
+        "dev_terms": args.dev_terms,
+        "expand_units": args.expand_units,
+        "expand_numbers": args.expand_numbers,
+        "pauses": args.pauses,
+        "tables": args.tables,
+        "strip_emoji": args.strip_emoji,
+        "terminal_stops": args.terminal_stops,
+        "locale": args.locale,
+    }
+
+
+def _resolve_normalized(text: str, args: argparse.Namespace) -> str | None:
+    """Normalize ``text`` for speech, or print a clean error and return None.
+
+    A missing or malformed ``--pronunciations`` file surfaces here as an
+    ``OSError`` / ``ValueError`` (``json.JSONDecodeError`` is a ``ValueError``);
+    turn it into a one-line ``[stackvox] …`` message rather than a traceback.
+    """
+    try:
+        return normalize_for_speech(text, **_normalize_kwargs(args))
+    except (OSError, ValueError) as exc:
+        print(f"[stackvox] {exc}", file=sys.stderr)
+        return None
+
+
 def _read_text(args: argparse.Namespace) -> str | None:
     """Resolve the text to speak from --file, the positional, or piped stdin.
 
@@ -177,6 +321,10 @@ def _cmd_speak(args: argparse.Namespace) -> int:
     if not text or not text.strip():
         print("Error: provide text or --file", file=sys.stderr)
         return 1
+    if getattr(args, "normalize", False):
+        text = _resolve_normalized(text, args)
+        if text is None:
+            return 1
     tts = Stackvox(voice=args.voice, speed=args.speed, lang=args.lang)
     if args.out:
         samples, sr = tts.synthesize(text)
@@ -192,6 +340,10 @@ def _cmd_say(args: argparse.Namespace) -> int:
     if not text or not text.strip():
         print("Error: provide text or --file", file=sys.stderr)
         return 1
+    if getattr(args, "normalize", False):
+        text = _resolve_normalized(text, args)
+        if text is None:
+            return 1
     ok, resp = daemon.say(text, voice=args.voice, speed=args.speed, lang=args.lang)
     if ok:
         return 0
@@ -206,6 +358,18 @@ def _cmd_say(args: argparse.Namespace) -> int:
     return 2
 
 
+def _cmd_normalize(args: argparse.Namespace) -> int:
+    text = _read_text(args)
+    if not text or not text.strip():
+        print("Error: provide text or --file", file=sys.stderr)
+        return 1
+    normalized = _resolve_normalized(text, args)
+    if normalized is None:
+        return 1
+    print(normalized)
+    return 0
+
+
 def _cmd_serve(args: argparse.Namespace) -> int:
     try:
         daemon.serve(voice=args.voice, speed=args.speed, lang=args.lang)
@@ -248,6 +412,24 @@ def _cmd_voices(args: argparse.Namespace) -> int:
     return 0
 
 
+def _cmd_paths(_: argparse.Namespace) -> int:
+    print(f"cache_dir:   {paths.cache_dir()}")
+    print(f"socket_path: {paths.socket_path()}")
+    print(f"pid_path:    {paths.pid_path()}")
+    return 0
+
+
+def _cmd_config(_: argparse.Namespace) -> int:
+    path = config.config_path()
+    present = "" if path.is_file() else " (not present; built-in defaults apply)"
+    defaults = config.load_defaults()
+    print(f"config_path: {path}{present}")
+    print(f"voice: {defaults.voice}")
+    print(f"speed: {defaults.speed}")
+    print(f"lang:  {defaults.lang}")
+    return 0
+
+
 def _cmd_welcome(_: argparse.Namespace) -> int:
     tts = Stackvox()
     tts.speak_sequence([{"text": text, "voice": voice, "lang": lang} for voice, lang, text in WELCOME_LINES])
@@ -336,11 +518,14 @@ def main() -> int:
     handlers = {
         "speak": _cmd_speak,
         "say": _cmd_say,
+        "normalize": _cmd_normalize,
         "serve": _cmd_serve,
         "stop": _cmd_stop,
         "status": _cmd_status,
         "voices": _cmd_voices,
         "welcome": _cmd_welcome,
+        "paths": _cmd_paths,
+        "config": _cmd_config,
         "completion": _cmd_completion,
         "install-helper": _cmd_install_helper,
     }

{stackvox-0.6.0 → stackvox-0.8.0}/stackvox/engine.py

@@ -22,6 +22,12 @@ DEFAULT_VOICE = "af_sarah"
 DEFAULT_SPEED = 1.0
 DEFAULT_LANG = "en-us"
 
+# Streaming playback smoothing: play a touch of silence before the first word so
+# output-device warm-up (notably a Bluetooth codec/profile switch) lands during
+# silence, and ramp the opening samples up from zero to avoid an onset click.
+_PRIME_SILENCE_SECONDS = 0.12
+_FADE_IN_SECONDS = 0.008
+
 _MODEL_URL = "https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files-v1.0/kokoro-v1.0.onnx"
 _VOICES_URL = "https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files-v1.0/voices-v1.0.bin"
 
@@ -82,6 +88,16 @@ def _split_sentences(text: str) -> list[str]:
     return [part.strip() for part in _SENTENCE_BOUNDARY.split(text.strip()) if part.strip()]
 
 
+def _fade_in(samples: np.ndarray, sample_rate: int) -> np.ndarray:
+    """Ramp the opening samples up from zero so playback doesn't begin with a click."""
+    ramp_len = min(len(samples), int(sample_rate * _FADE_IN_SECONDS))
+    if ramp_len <= 0:
+        return samples
+    faded = np.array(samples, dtype="float32", copy=True)
+    faded[:ramp_len] *= np.linspace(0.0, 1.0, ramp_len, dtype="float32")
+    return faded
+
+
 class Stackvox:
     """Reusable TTS engine. Load the model once, speak many times.
 
@@ -104,6 +120,8 @@ class Stackvox:
         self._kokoro = Kokoro(str(model_path), str(voices_path))
         self._stop_event = threading.Event()
         self._play_thread: threading.Thread | None = None
+        self._stream: sd.OutputStream | None = None
+        self._stream_lock = threading.Lock()
 
     def synthesize(
         self,
@@ -210,6 +228,7 @@ class Stackvox:
         producer = threading.Thread(target=_produce, daemon=True, name="stackvox-synth")
         producer.start()
         error: Exception | None = None
+        stream: sd.OutputStream | None = None
         try:
             while not stop_event.is_set():
                 item = chunks.get()
@@ -219,9 +238,36 @@ class Stackvox:
                     error = item
                     break
                 samples, sample_rate = item
-                sd.play(samples, sample_rate)
-                sd.wait()
+                samples = np.ascontiguousarray(samples, dtype="float32")
+                if stream is None:
+                    # One stream for the whole utterance: writing chunks into it is
+                    # gapless and avoids the per-sentence open/close that crackles at
+                    # the start. Prime with silence so device warm-up (e.g. a
+                    # Bluetooth codec switch) lands before the first word, and fade
+                    # the opening samples in to avoid an onset click.
+                    stream = sd.OutputStream(samplerate=sample_rate, channels=1, dtype="float32")
+                    stream.start()
+                    with self._stream_lock:
+                        self._stream = stream
+                    prime = int(sample_rate * _PRIME_SILENCE_SECONDS)
+                    if prime > 0:
+                        stream.write(np.zeros(prime, dtype="float32"))
+                    samples = _fade_in(samples, sample_rate)
+                try:
+                    stream.write(samples)
+                except Exception:
+                    if stop_event.is_set():
+                        break  # aborted by stop(); expected
+                    raise
         finally:
+            with self._stream_lock:
+                self._stream = None
+            if stream is not None:
+                try:
+                    stream.stop()
+                    stream.close()
+                except Exception:
+                    logger.debug("stream teardown failed", exc_info=True)
             # Stop the producer and keep draining so it can never stay parked on
             # a full queue — that's what guarantees the synth thread always exits.
             stop_event.set()
@@ -236,6 +282,12 @@ class Stackvox:
     def stop(self) -> None:
         """Stop any in-progress playback started with blocking=False."""
         self._stop_event.set()
+        with self._stream_lock:
+            if self._stream is not None:
+                try:
+                    self._stream.abort()
+                except Exception:
+                    logger.debug("stream abort failed", exc_info=True)
         sd.stop()
 
     def voices(self) -> list[str]:

{stackvox-0.6.0 → stackvox-0.8.0}/stackvox/text.py

@@ -106,6 +106,28 @@ def apply_pronunciations(text: str, mapping: dict[str, str] | None) -> str:
     return text
 
 
+# Dev acronyms/terms espeak mispronounces — it reads them as a word ("CLI" ->
+# "kligh", "AWS" -> "awz", "URI" -> "yuri") instead of spelling them out.
+# Applied by default (dev_terms=True), whole-word and case-insensitive, so
+# lowercase "cli" is fixed too. Compound keys (ci/cd) precede their parts (ci)
+# so the specific form wins. Only terms espeak gets WRONG are here — API, URL,
+# JSON, YAML, HTTP, CRUD, nginx, etc. already voice correctly and are left alone.
+_DEV_PRONUNCIATIONS: dict[str, str] = {
+    "ci/cd": "C I C D",
+    "cli": "C L I",
+    "ci": "C I",
+    "ide": "I D E",
+    "aws": "A W S",
+    "uri": "U R I",
+    "iam": "I A M",
+    "saas": "sass",
+    "paas": "pass",
+    "tui": "T U I",
+    "postgresql": "postgres",
+    "kubectl": "kube control",
+}
+
+
 # --------------------------------------------------------------------------- #
 # Pauses                                                                      #
 # --------------------------------------------------------------------------- #
@@ -274,6 +296,7 @@ def normalize_for_speech(
     *,
     markdown: bool = True,
     pronunciations: dict[str, str] | None = None,
+    dev_terms: bool = True,
     expand_units: bool = True,
     expand_numbers: bool = True,
     pauses: bool = True,
@@ -285,6 +308,11 @@ def normalize_for_speech(
     """Normalize ``text`` into speakable prose. Returns paragraphs joined by
     newlines. See ``docs/speech-normalization.md`` for the full contract."""
     expand_units_flag, expand_numbers_flag = expand_units, expand_numbers
+    # Built-in dev-term fixes first; caller-supplied pronunciations override them
+    # (keyed case-insensitively, so a caller's "CLI" beats the default "cli").
+    effective_pronunciations: dict[str, str] = dict(_DEV_PRONUNCIATIONS) if dev_terms else {}
+    for written, spoken in (pronunciations or {}).items():
+        effective_pronunciations[written.lower()] = spoken
 
     if markdown:
         paragraphs = markdown_to_paragraphs(text, tables=tables, strip_emoji_flag=strip_emoji)
@@ -298,7 +326,7 @@ def normalize_for_speech(
     for para in paragraphs:
         shaped = _shape_paragraph(
             para,
-            pronunciations=pronunciations,
+            pronunciations=effective_pronunciations,
             expand_units_flag=expand_units_flag,
             expand_numbers_flag=expand_numbers_flag,
             pauses_flag=pauses,

{stackvox-0.6.0 → stackvox-0.8.0/stackvox.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: stackvox
-Version: 0.6.0
+Version: 0.8.0
 Summary: Offline TTS library using Kokoro-82M
 Author-email: "StackOne Technologies Ltd." <support@stackone.com>
 License-Expression: Apache-2.0
@@ -88,8 +88,23 @@ stackvox welcome                    # multilingual welcome (6 languages)
 stackvox voices                     # list all voice ids
 echo "from a pipe" | stackvox       # piped stdin works for speak/say
 stackvox speak --file message.txt   # read a whole file
+stackvox paths                      # print resolved cache / socket / pid paths
+stackvox config                     # print resolved default voice/speed/lang
 ```
 
+### Speech normalization
+
+`normalize_for_speech` (Markdown → speakable prose: strip formatting, expand units/numbers, shape pauses, apply a pronunciation dictionary) is available from the shell — no Python import required:
+
+```bash
+stackvox normalize --file post.md               # print normalized text (pipe / debug)
+stackvox normalize --tables csv --strip-emoji < response.md
+stackvox speak --normalize --file post.md       # normalize, then synthesize
+stackvox say   --normalize "**Shipped** 1,198.9 MPG"   # normalize, then send to daemon
+```
+
+`--normalize` is off by default on `speak`/`say`, so notification phrases synthesize exactly as before. The tuning flags mirror the library defaults (all on except emoji-strip): `--no-markdown`, `--no-expand-units`, `--no-expand-numbers`, `--no-pauses`, `--tables {drop,csv}`, `--strip-emoji`, `--no-terminal-stops`, `--locale`, and a `--pronunciations FILE` (a JSON `{"written": "spoken"}` map, applied whole-word and case-insensitively).
+
 Bash completion:
 
 ```bash

{stackvox-0.6.0 → stackvox-0.8.0}/tests/test_cli.py

@@ -121,6 +121,26 @@ def _ns(**kwargs):
     return argparse.Namespace(**kwargs)
 
 
+def _norm_ns(text=None, file=None, **overrides):
+    """Namespace carrying the normalize flags at their library defaults."""
+    base = dict(
+        text=text,
+        file=file,
+        markdown=True,
+        pronunciations=None,
+        dev_terms=True,
+        expand_units=True,
+        expand_numbers=True,
+        pauses=True,
+        tables="drop",
+        strip_emoji=False,
+        terminal_stops=True,
+        locale="en-GB",
+    )
+    base.update(overrides)
+    return _ns(**base)
+
+
 class TestCmdSpeak:
     def test_with_text_calls_engine_speak(self, fake_stackvox):
         rc = cli._cmd_speak(_ns(voice="af_sarah", speed=1.0, lang="en-us", text="hello", out=None))
@@ -149,6 +169,18 @@ class TestCmdSpeak:
         assert "provide text" in capsys.readouterr().err
         fake_stackvox.return_value.speak.assert_not_called()
 
+    def test_normalize_transforms_text_before_engine(self, fake_stackvox):
+        args = _norm_ns(text="# Heading", voice="af_sarah", speed=1.0, lang="en-us", out=None, normalize=True)
+        rc = cli._cmd_speak(args)
+        assert rc == 0
+        fake_stackvox.return_value.speak.assert_called_once_with("Heading.")
+
+    def test_no_normalize_leaves_text_untouched(self, fake_stackvox):
+        # `normalize` absent from the namespace mirrors the flag being off.
+        rc = cli._cmd_speak(_ns(voice="af_sarah", speed=1.0, lang="en-us", text="# Heading", out=None))
+        assert rc == 0
+        fake_stackvox.return_value.speak.assert_called_once_with("# Heading")
+
 
 class TestCmdSay:
     def test_returns_zero_when_daemon_accepts(self, mocker):
@@ -182,6 +214,15 @@ class TestCmdSay:
         rc = cli._cmd_say(_ns(voice="af_sarah", speed=1.0, lang="en-us", text="", fallback_say=False))
         assert rc == 1
 
+    def test_normalize_transforms_text_before_daemon(self, mocker):
+        say = mocker.patch.object(cli.daemon, "say", return_value=(True, "ok"))
+        args = _norm_ns(
+            text="# Heading", voice="af_sarah", speed=1.0, lang="en-us", fallback_say=False, normalize=True
+        )
+        rc = cli._cmd_say(args)
+        assert rc == 0
+        assert say.call_args.args[0] == "Heading."
+
 
 class TestCmdServe:
     def test_propagates_serve_args(self, mocker):
@@ -353,3 +394,112 @@ class TestCmdInstallHelper:
         mocker.patch.dict("os.environ", {"PATH": f"{prefix}:/usr/bin"}, clear=False)
         cli._cmd_install_helper(_ns(prefix=prefix))
         assert "not on your PATH" not in capsys.readouterr().err
+
+
+class TestCmdNormalize:
+    def test_markdown_reduces_to_prose_with_terminal_stops(self, capsys):
+        rc = cli._cmd_normalize(_norm_ns(text="# Heading\n\nBody"))
+        assert rc == 0
+        assert capsys.readouterr().out.splitlines() == ["Heading.", "Body."]
+
+    def test_number_and_decimal_expansion(self, capsys):
+        rc = cli._cmd_normalize(_norm_ns(text="It was 1,198.9 total"))
+        assert rc == 0
+        assert capsys.readouterr().out.strip() == "It was 1198 point 9 total."
+
+    def test_pronunciations_file_is_applied(self, capsys, tmp_path):
+        pron = tmp_path / "dict.json"
+        pron.write_text('{"Redis": "ree-diss"}', encoding="utf-8")
+        rc = cli._cmd_normalize(_norm_ns(text="Redis is fast", pronunciations=pron))
+        assert rc == 0
+        assert capsys.readouterr().out.strip() == "ree-diss is fast."
+
+    def test_blank_input_returns_one(self, capsys):
+        rc = cli._cmd_normalize(_norm_ns(text="   "))
+        assert rc == 1
+        assert "provide text" in capsys.readouterr().err
+
+    def test_bad_pronunciations_file_errors_cleanly(self, capsys, tmp_path):
+        bad = tmp_path / "bad.json"
+        bad.write_text('["not", "a", "map"]', encoding="utf-8")
+        rc = cli._cmd_normalize(_norm_ns(text="hi", pronunciations=bad))
+        assert rc == 1
+        assert "string -> string" in capsys.readouterr().err
+
+
+class TestLoadPronunciations:
+    def test_none_path_returns_none(self):
+        assert cli._load_pronunciations(None) is None
+
+    def test_valid_json_returns_mapping(self, tmp_path):
+        path = tmp_path / "d.json"
+        path.write_text('{"CLI": "see ell eye"}', encoding="utf-8")
+        assert cli._load_pronunciations(path) == {"CLI": "see ell eye"}
+
+    def test_non_object_json_raises(self, tmp_path):
+        path = tmp_path / "d.json"
+        path.write_text('["not", "a", "map"]', encoding="utf-8")
+        with pytest.raises(ValueError, match="string -> string"):
+            cli._load_pronunciations(path)
+
+    def test_non_string_values_raise(self, tmp_path):
+        path = tmp_path / "d.json"
+        path.write_text('{"n": 3}', encoding="utf-8")
+        with pytest.raises(ValueError, match="string -> string"):
+            cli._load_pronunciations(path)
+
+
+class TestCmdPaths:
+    def test_prints_the_three_paths(self, capsys):
+        rc = cli._cmd_paths(_ns())
+        assert rc == 0
+        out = capsys.readouterr().out
+        assert "cache_dir:" in out
+        assert "socket_path:" in out
+        assert "pid_path:" in out
+
+
+class TestCmdConfig:
+    def test_prints_resolved_defaults_and_path(self, mocker, capsys, tmp_path):
+        cfg = tmp_path / "config.toml"  # absent → "not present" note
+        mocker.patch.object(cli.config, "config_path", return_value=cfg)
+        mocker.patch.object(
+            cli.config,
+            "load_defaults",
+            return_value=cli.config.Defaults(voice="bf_emma", speed=1.1, lang="en-gb"),
+        )
+        rc = cli._cmd_config(_ns())
+        assert rc == 0
+        out = capsys.readouterr().out
+        assert "not present" in out
+        assert "bf_emma" in out
+        assert "1.1" in out
+        assert "en-gb" in out
+
+
+class TestNormalizeParsing:
+    def test_normalize_subcommand_routes(self, mocker):
+        norm = mocker.patch.object(cli, "_cmd_normalize", return_value=0)
+        mocker.patch.object(cli.sys, "argv", ["stackvox", "normalize", "--tables", "csv", "hi"])
+        assert cli.main() == 0
+        args = norm.call_args.args[0]
+        assert args.cmd == "normalize"
+        assert args.tables == "csv"
+        assert args.text == "hi"
+
+    def test_speak_normalize_switch_parses(self, mocker):
+        speak = mocker.patch.object(cli, "_cmd_speak", return_value=0)
+        mocker.patch.object(
+            cli.sys, "argv", ["stackvox", "speak", "--normalize", "--strip-emoji", "--no-pauses", "hi"]
+        )
+        assert cli.main() == 0
+        args = speak.call_args.args[0]
+        assert args.normalize is True
+        assert args.strip_emoji is True
+        assert args.pauses is False
+
+    def test_paths_and_config_route(self, mocker):
+        paths_handler = mocker.patch.object(cli, "_cmd_paths", return_value=0)
+        mocker.patch.object(cli.sys, "argv", ["stackvox", "paths"])
+        assert cli.main() == 0
+        assert paths_handler.called

{stackvox-0.6.0 → stackvox-0.8.0}/tests/test_engine.py

@@ -35,7 +35,8 @@ def fake_kokoro(mocker, fake_ensure_models):
 
 @pytest.fixture
 def fake_audio(mocker):
-    """Mock the sounddevice surface used by Stackvox."""
+    """Mock the sounddevice surface used by Stackvox (streaming + speak_sequence)."""
+    mocker.patch.object(engine.sd, "OutputStream")
     mocker.patch.object(engine.sd, "play")
     mocker.patch.object(engine.sd, "wait")
     mocker.patch.object(engine.sd, "stop")
@@ -105,15 +106,35 @@ class TestSpeak:
 
         engine.Stackvox().speak("One sentence. Two sentence. Three here.")
 
-        # One synth + play cycle per sentence — that's what enables early audio.
+        # One synth per sentence (early audio), all streamed through a single
+        # output stream: a prime-silence write plus one write per sentence.
         assert fake_kokoro.return_value.create.call_count == 3
-        assert engine.sd.play.call_count == 3
-        assert engine.sd.wait.call_count == 3
+        engine.sd.OutputStream.assert_called_once()
+        stream = engine.sd.OutputStream.return_value
+        assert stream.write.call_count == 4  # 1 prime + 3 sentences
+        stream.close.assert_called_once()
 
     def test_single_sentence_plays_once(self, fake_kokoro, fake_audio):
         fake_kokoro.return_value.create.return_value = (np.zeros(10, dtype=np.float32), 24000)
         engine.Stackvox().speak("Just one sentence with no terminal punctuation")
-        assert engine.sd.play.call_count == 1
+        engine.sd.OutputStream.assert_called_once()
+        assert engine.sd.OutputStream.return_value.write.call_count == 2  # prime + 1 sentence
+
+    def test_primes_with_silence_before_first_audio(self, fake_kokoro, fake_audio):
+        # The lead-in silence lets the output device (e.g. Bluetooth) warm up
+        # before the first word, rather than crackling into it.
+        fake_kokoro.return_value.create.return_value = (np.ones(10, dtype=np.float32), 24000)
+        engine.Stackvox().speak("hello")
+        prime = engine.sd.OutputStream.return_value.write.call_args_list[0].args[0]
+        assert not prime.any()
+        assert len(prime) == int(24000 * engine._PRIME_SILENCE_SECONDS)
+
+    def test_first_audio_fades_in_from_zero(self, fake_kokoro, fake_audio):
+        fake_kokoro.return_value.create.return_value = (np.ones(1000, dtype=np.float32), 24000)
+        engine.Stackvox().speak("hello")
+        first_audio = engine.sd.OutputStream.return_value.write.call_args_list[1].args[0]
+        assert first_audio[0] == 0.0  # ramp starts at silence
+        assert first_audio[-1] == 1.0  # ramp completes, body untouched
 
     def test_non_blocking_plays_in_background_thread(self, fake_kokoro, fake_audio):
         fake_kokoro.return_value.create.return_value = (np.zeros(10, dtype=np.float32), 24000)
@@ -123,7 +144,7 @@ class TestSpeak:
 
         assert tts._play_thread is not None
         tts._play_thread.join(timeout=5)
-        engine.sd.play.assert_called()
+        engine.sd.OutputStream.assert_called()
 
     def test_passes_engine_defaults_to_synthesis(self, fake_kokoro, fake_audio):
         fake_kokoro.return_value.create.return_value = (np.zeros(4, dtype=np.float32), 24000)
@@ -177,7 +198,7 @@ class TestSpeak:
         tts = engine.Stackvox()
         tts.stop()
         tts.speak("hello world")
-        engine.sd.play.assert_called()
+        engine.sd.OutputStream.assert_called()
 
 
 class TestSplitSentences:

{stackvox-0.6.0 → stackvox-0.8.0}/tests/test_text.py

@@ -195,3 +195,28 @@ def test_speaklast_style_normalization():
 def test_plain_text_mode_splits_paragraphs():
     out = normalize_for_speech("First para.\n\nSecond para.", markdown=False)
     assert out == "First para.\nSecond para."
+
+
+def test_dev_acronyms_are_spelled_out():
+    assert normalize_for_speech("Use the CLI.", markdown=False) == "Use the C L I."
+    assert "C L I" in normalize_for_speech("the cli tool", markdown=False)  # lowercase too
+    assert "A W S" in normalize_for_speech("deploy to AWS", markdown=False)
+    assert "U R I" in normalize_for_speech("parse the URI", markdown=False)
+    assert "C I C D" in normalize_for_speech("the CI/CD pipeline", markdown=False)
+
+
+def test_dev_terms_leave_correctly_voiced_acronyms_alone():
+    out = normalize_for_speech("Send JSON over HTTP to the API.", markdown=False)
+    assert "JSON" in out
+    assert "HTTP" in out
+    assert "API" in out
+
+
+def test_dev_terms_can_be_disabled():
+    assert normalize_for_speech("Use the CLI.", markdown=False, dev_terms=False) == "Use the CLI."
+
+
+def test_caller_pronunciations_override_dev_terms():
+    out = normalize_for_speech("Use the CLI.", markdown=False, pronunciations={"CLI": "command line"})
+    assert "command line" in out
+    assert "C L I" not in out