ninada 4.0.0__tar.gz

ninada-4.0.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Hemanth HM <hemanth@h3manth.com> (https://h3manth.com)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

ninada-4.0.0/PKG-INFO

@@ -0,0 +1,134 @@
+Metadata-Version: 2.4
+Name: ninada
+Version: 4.0.0
+Summary: Ninada (निनाद) — Streaming 24 kHz Native Per-Language Studio VITS neural text-to-speech for Classical Sanskrit & 5 Indian languages (6 dedicated ~28 MB W8A32+FP16 ONNX specialists for Browser WASM SIMD, Node.js, and Python).
+Author-email: Hemanth HM <hemanth@h3manth.com>
+License: MIT
+Project-URL: Homepage, https://hemanth.github.io/ninada/
+Project-URL: Repository, https://github.com/hemanth/ninada
+Project-URL: Model, https://huggingface.co/gnumanth/ninada
+Project-URL: Dataset, https://huggingface.co/datasets/gnumanth/ninada-sanskrit-indic
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: neural
+Requires-Dist: onnxruntime>=1.18.0; extra == "neural"
+Requires-Dist: numpy>=1.24.0; extra == "neural"
+Requires-Dist: soundfile>=0.12.1; extra == "neural"
+Provides-Extra: train
+Requires-Dist: torch>=2.2.0; extra == "train"
+Requires-Dist: torchaudio>=2.2.0; extra == "train"
+Requires-Dist: transformers>=4.44.0; extra == "train"
+Requires-Dist: safetensors>=0.4.3; extra == "train"
+Requires-Dist: onnx>=1.16.0; extra == "train"
+Requires-Dist: onnxruntime>=1.18.0; extra == "train"
+Requires-Dist: huggingface_hub>=0.24.0; extra == "train"
+Requires-Dist: pyarrow>=15.0.0; extra == "train"
+Requires-Dist: scipy>=1.11.0; extra == "train"
+Requires-Dist: soundfile>=0.12.1; extra == "train"
+Provides-Extra: dev
+Requires-Dist: pytest>=8.0.0; extra == "dev"
+Dynamic: license-file
+
+# ninada
+
+Stream 24,000 Hz studio speech for Classical Sanskrit and five Indian languages (`sa`, `kn`, `hi`, `ta`, `te`, `ml`) in the browser, Node.js, and Python using `27.9 MB` native per-language VITS ONNX specialists (Male & Female) with zero cross-language accent bleed.
+
+```bash
+npm install ninada onnxruntime-web
+```
+
+```bash
+pip install ninada
+```
+
+## Quick start
+
+```js
+import ninada from 'ninada';
+
+const voice = await ninada('kannada', {
+  models: {
+    kn: 'https://huggingface.co/gnumanth/ninada/resolve/main/onnx/kn.onnx',
+    sa: 'https://huggingface.co/gnumanth/ninada/resolve/main/onnx/sa.onnx',
+  },
+  cache: true,
+});
+
+for await (const chunk of voice.stream('ನಮಸ್ಕಾರ, ಇಂದು ಬೆಂಗಳೂರಿನಲ್ಲಿ ಹವಾಮಾನ ತುಂಬಾ ಚೆನ್ನಾಗಿದೆ.')) {
+  playPcm24k(chunk.pcm, chunk.sampleRate); // Float32Array @ 24,000 Hz
+}
+
+const wav = await voice.speak('ಕನ್ನಡವೇ ಸತ್ಯ ಕನ್ನಡವೇ ನಿತ್ಯ', { lang: 'kn', voice: 'female' });
+await voice.dispose();
+```
+
+`voice.stream()` yields `24,000 Hz` PCM clauses as they synthesize with `0` `AudioWorklet` underruns and `CacheStorage` + `IndexedDB` persistence (`0` refetches on reload). `voice.speak()` returns a complete `24 kHz` RIFF WAV buffer.
+
+## Python
+
+```python
+from ninada import Ninada
+
+tts = Ninada(lang="sa")
+for chunk in tts.stream("प्रज्ञापूर्वकव्यापारवत्त्वं जीवत्वम् । यथार्थज्ञानं प्रमाणम् ॥"):
+    play(chunk["pcm"], chunk["sample_rate"])  # float32 @ 24,000 Hz (~0.197x CPU RTF)
+
+wav_bytes = tts.speak("सत्यमेव जयते नानृतम्", lang="sa", voice="female")
+open("satyam.wav", "wb").write(wav_bytes)
+```
+
+`Ninada` loads dedicated `27.9 MB` W8A32+FP16 ONNX specialists (`onnx/{sa,kn,hi,ta,te,ml}.onnx` and `onnx/{sa,kn,hi,ta,te,ml}_female.onnx`), each trained exclusively on its own language's `ai4bharat/Rasa` single-speaker `24 kHz` studio recordings and native Unicode orthography.
+
+## CLI
+
+```bash
+npx ninada "ಕನ್ನಡವೇ ಸತ್ಯ ಕನ್ನಡವೇ ನಿತ್ಯ" -l kn -o kannada.wav
+ninada "सत्यमेव जयते नानृतम्" -l sa -o satyam.wav
+```
+
+Runs `24 kHz` neural synthesis when ONNX weights are available and falls back automatically to a `0.0 MB` zero-dependency Cascade Klatt-Akṣara synthesizer offline.
+
+## Native Script Orthography & Prosody
+
+```js
+import { studioTextToIds, textToIds } from 'ninada';
+
+const { studioIds, lang } = studioTextToIds('ವಾಙ್ಮಯ ತಪಸ್ಸು ಮತ್ತು ಜ್ಞಾನ ವಿಕಾಸ ಅತ್ಯಂತ ಮುಖ್ಯವಾದವು.', 'kn');
+const { aksharas } = textToIds('धर्मक्षेत्रे कुरुक्षेत्रे समवेता युयुत्सवः ।', 'sa');
+```
+
+`studioTextToIds()` tokenizes directly in each language's native Unicode character table (`sa` 144 tokens, `kn` 85, `hi` 120, `ta` 61, `te` 110, `ml` 101) — preserving Classical Sanskrit vocalic `ṛ`/`ṝ`/`ḷ`, visarga sandhi, Kannada `ಙ`/`ಃ` allophony, Tamil `ழ`/`ற`/`ன`, and Malayalam atomic chillus (`ൺ ൻ ർ ൽ ൾ ൿ`) feeding a `24 kHz` `RelativePositionTransformer + StochasticDurationPredictor + HiFi-GAN v1` studio generator.
+
+## Benchmarks (`N = 240` Held-Out 24 kHz Studio Utterances, `B = 1,000` Bootstrap 95% CIs)
+
+| System | Rate | Footprint (↓) | Coverage (↑) | TTFA p50 (↓) | MCD 0–8 kHz (dB, ↓) | 8–12 kHz HF Err (dB, ↓) | Dur MAE (↓) |
+|---|---:|---:|---:|---:|---:|---:|---:|
+| Base `facebook/mms-tts-hin` | 16 kHz | 138.8 MB | 35.0% | 1.0 ms* | `7.97 [7.69, 8.24]` | 66.24 dB | 3.21s |
+| Meta MMS 5-Specialist Fleet (`5x` VITS) | 16 kHz | 694.0 MB | 100.0% | 60.2 ms | `5.09 [5.03, 5.15]` | 65.90 dB | 1.34s |
+| `ninada` v4.0 Untuned Warm-Start (`step=0`) | 24 kHz | 112.0 MB | 100.0% | 189.2 ms | `4.60 [4.54, 4.65]` | 11.01 dB | 0.95s |
+| **`ninada` v4.0 Native Studio FP32** | **24 kHz** | 112.0 MB | **100.0%** | **45.3 ms** | **`3.40 [3.37, 3.43]`** | **6.34 dB** | **0.57s** |
+| **`ninada` v4.0 Native Browser ONNX** | **24 kHz** | **27.9 MB** | **100.0%** | **372.7 ms** | **`3.39 [3.36, 3.43]`** | **6.37 dB** | **0.56s** |
+| `ninada` Tier-2 Klatt Fallback | 24 kHz | **0.0 MB** | **100.0%** | **104.0 ms** | `10.38 [10.26, 10.50]` | 41.94 dB | 0.98s |
+
+Per-language held-out `MCD 0–8 kHz (dB, ↓)`: Sanskrit **`3.51`** (vs `5.32` MMS), Hindi **`3.37`** (vs `4.85` MMS), Kannada **`3.35`** (vs `5.04` MMS), Tamil **`3.21`** (vs `4.87` MMS), Telugu **`3.59`** (vs `5.57` MMS), Malayalam **`3.34`** (vs `4.88` MMS). Full bootstrap CIs and parity telemetry (`0.61 dB` parity MCD, `r = 0.9999`) in [`bench/SCIENCE_REPORT.md`](./bench/SCIENCE_REPORT.md) and [`bench/results.json`](./bench/results.json).
+
+## Reproduce & Verify
+
+```bash
+PYTHONPATH=python/src:. python training/train_native_specialists_gcp.py --work-dir data/v4 --out-dir artifacts/ninada_v4
+PYTHONPATH=python/src:. python bench/check_bench_thresholds.py
+pytest -q && node verify-browser.mjs
+```
+
+Trains all 12 native `24 kHz` Male/Female studio VITS specialists, evaluates `240` held-out studio clips, and verifies `0` `AudioWorklet` underruns and `0` cached network refetches in headless Chromium.
+
+## Links
+
+- [Live Browser Voice Studio](https://hemanth.github.io/ninada/)
+- [Hugging Face Model (`gnumanth/ninada`)](https://huggingface.co/gnumanth/ninada)
+- [Hugging Face Dataset (`gnumanth/ninada-sanskrit-indic`)](https://huggingface.co/datasets/gnumanth/ninada-sanskrit-indic)
+
+## License
+
+MIT © [Hemanth.HM](https://h3manth.com)

ninada-4.0.0/README.md

@@ -0,0 +1,102 @@
+# ninada
+
+Stream 24,000 Hz studio speech for Classical Sanskrit and five Indian languages (`sa`, `kn`, `hi`, `ta`, `te`, `ml`) in the browser, Node.js, and Python using `27.9 MB` native per-language VITS ONNX specialists (Male & Female) with zero cross-language accent bleed.
+
+```bash
+npm install ninada onnxruntime-web
+```
+
+```bash
+pip install ninada
+```
+
+## Quick start
+
+```js
+import ninada from 'ninada';
+
+const voice = await ninada('kannada', {
+  models: {
+    kn: 'https://huggingface.co/gnumanth/ninada/resolve/main/onnx/kn.onnx',
+    sa: 'https://huggingface.co/gnumanth/ninada/resolve/main/onnx/sa.onnx',
+  },
+  cache: true,
+});
+
+for await (const chunk of voice.stream('ನಮಸ್ಕಾರ, ಇಂದು ಬೆಂಗಳೂರಿನಲ್ಲಿ ಹವಾಮಾನ ತುಂಬಾ ಚೆನ್ನಾಗಿದೆ.')) {
+  playPcm24k(chunk.pcm, chunk.sampleRate); // Float32Array @ 24,000 Hz
+}
+
+const wav = await voice.speak('ಕನ್ನಡವೇ ಸತ್ಯ ಕನ್ನಡವೇ ನಿತ್ಯ', { lang: 'kn', voice: 'female' });
+await voice.dispose();
+```
+
+`voice.stream()` yields `24,000 Hz` PCM clauses as they synthesize with `0` `AudioWorklet` underruns and `CacheStorage` + `IndexedDB` persistence (`0` refetches on reload). `voice.speak()` returns a complete `24 kHz` RIFF WAV buffer.
+
+## Python
+
+```python
+from ninada import Ninada
+
+tts = Ninada(lang="sa")
+for chunk in tts.stream("प्रज्ञापूर्वकव्यापारवत्त्वं जीवत्वम् । यथार्थज्ञानं प्रमाणम् ॥"):
+    play(chunk["pcm"], chunk["sample_rate"])  # float32 @ 24,000 Hz (~0.197x CPU RTF)
+
+wav_bytes = tts.speak("सत्यमेव जयते नानृतम्", lang="sa", voice="female")
+open("satyam.wav", "wb").write(wav_bytes)
+```
+
+`Ninada` loads dedicated `27.9 MB` W8A32+FP16 ONNX specialists (`onnx/{sa,kn,hi,ta,te,ml}.onnx` and `onnx/{sa,kn,hi,ta,te,ml}_female.onnx`), each trained exclusively on its own language's `ai4bharat/Rasa` single-speaker `24 kHz` studio recordings and native Unicode orthography.
+
+## CLI
+
+```bash
+npx ninada "ಕನ್ನಡವೇ ಸತ್ಯ ಕನ್ನಡವೇ ನಿತ್ಯ" -l kn -o kannada.wav
+ninada "सत्यमेव जयते नानृतम्" -l sa -o satyam.wav
+```
+
+Runs `24 kHz` neural synthesis when ONNX weights are available and falls back automatically to a `0.0 MB` zero-dependency Cascade Klatt-Akṣara synthesizer offline.
+
+## Native Script Orthography & Prosody
+
+```js
+import { studioTextToIds, textToIds } from 'ninada';
+
+const { studioIds, lang } = studioTextToIds('ವಾಙ್ಮಯ ತಪಸ್ಸು ಮತ್ತು ಜ್ಞಾನ ವಿಕಾಸ ಅತ್ಯಂತ ಮುಖ್ಯವಾದವು.', 'kn');
+const { aksharas } = textToIds('धर्मक्षेत्रे कुरुक्षेत्रे समवेता युयुत्सवः ।', 'sa');
+```
+
+`studioTextToIds()` tokenizes directly in each language's native Unicode character table (`sa` 144 tokens, `kn` 85, `hi` 120, `ta` 61, `te` 110, `ml` 101) — preserving Classical Sanskrit vocalic `ṛ`/`ṝ`/`ḷ`, visarga sandhi, Kannada `ಙ`/`ಃ` allophony, Tamil `ழ`/`ற`/`ன`, and Malayalam atomic chillus (`ൺ ൻ ർ ൽ ൾ ൿ`) feeding a `24 kHz` `RelativePositionTransformer + StochasticDurationPredictor + HiFi-GAN v1` studio generator.
+
+## Benchmarks (`N = 240` Held-Out 24 kHz Studio Utterances, `B = 1,000` Bootstrap 95% CIs)
+
+| System | Rate | Footprint (↓) | Coverage (↑) | TTFA p50 (↓) | MCD 0–8 kHz (dB, ↓) | 8–12 kHz HF Err (dB, ↓) | Dur MAE (↓) |
+|---|---:|---:|---:|---:|---:|---:|---:|
+| Base `facebook/mms-tts-hin` | 16 kHz | 138.8 MB | 35.0% | 1.0 ms* | `7.97 [7.69, 8.24]` | 66.24 dB | 3.21s |
+| Meta MMS 5-Specialist Fleet (`5x` VITS) | 16 kHz | 694.0 MB | 100.0% | 60.2 ms | `5.09 [5.03, 5.15]` | 65.90 dB | 1.34s |
+| `ninada` v4.0 Untuned Warm-Start (`step=0`) | 24 kHz | 112.0 MB | 100.0% | 189.2 ms | `4.60 [4.54, 4.65]` | 11.01 dB | 0.95s |
+| **`ninada` v4.0 Native Studio FP32** | **24 kHz** | 112.0 MB | **100.0%** | **45.3 ms** | **`3.40 [3.37, 3.43]`** | **6.34 dB** | **0.57s** |
+| **`ninada` v4.0 Native Browser ONNX** | **24 kHz** | **27.9 MB** | **100.0%** | **372.7 ms** | **`3.39 [3.36, 3.43]`** | **6.37 dB** | **0.56s** |
+| `ninada` Tier-2 Klatt Fallback | 24 kHz | **0.0 MB** | **100.0%** | **104.0 ms** | `10.38 [10.26, 10.50]` | 41.94 dB | 0.98s |
+
+Per-language held-out `MCD 0–8 kHz (dB, ↓)`: Sanskrit **`3.51`** (vs `5.32` MMS), Hindi **`3.37`** (vs `4.85` MMS), Kannada **`3.35`** (vs `5.04` MMS), Tamil **`3.21`** (vs `4.87` MMS), Telugu **`3.59`** (vs `5.57` MMS), Malayalam **`3.34`** (vs `4.88` MMS). Full bootstrap CIs and parity telemetry (`0.61 dB` parity MCD, `r = 0.9999`) in [`bench/SCIENCE_REPORT.md`](./bench/SCIENCE_REPORT.md) and [`bench/results.json`](./bench/results.json).
+
+## Reproduce & Verify
+
+```bash
+PYTHONPATH=python/src:. python training/train_native_specialists_gcp.py --work-dir data/v4 --out-dir artifacts/ninada_v4
+PYTHONPATH=python/src:. python bench/check_bench_thresholds.py
+pytest -q && node verify-browser.mjs
+```
+
+Trains all 12 native `24 kHz` Male/Female studio VITS specialists, evaluates `240` held-out studio clips, and verifies `0` `AudioWorklet` underruns and `0` cached network refetches in headless Chromium.
+
+## Links
+
+- [Live Browser Voice Studio](https://hemanth.github.io/ninada/)
+- [Hugging Face Model (`gnumanth/ninada`)](https://huggingface.co/gnumanth/ninada)
+- [Hugging Face Dataset (`gnumanth/ninada-sanskrit-indic`)](https://huggingface.co/datasets/gnumanth/ninada-sanskrit-indic)
+
+## License
+
+MIT © [Hemanth.HM](https://h3manth.com)

ninada-4.0.0/pyproject.toml

@@ -0,0 +1,53 @@
+[build-system]
+requires = ["setuptools>=68.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "ninada"
+version = "4.0.0"
+description = "Ninada (निनाद) — Streaming 24 kHz Native Per-Language Studio VITS neural text-to-speech for Classical Sanskrit & 5 Indian languages (6 dedicated ~28 MB W8A32+FP16 ONNX specialists for Browser WASM SIMD, Node.js, and Python)."
+readme = "README.md"
+license = { text = "MIT" }
+authors = [
+  { name = "Hemanth HM", email = "hemanth@h3manth.com" }
+]
+requires-python = ">=3.10"
+dependencies = []
+
+[project.urls]
+Homepage = "https://hemanth.github.io/ninada/"
+Repository = "https://github.com/hemanth/ninada"
+Model = "https://huggingface.co/gnumanth/ninada"
+Dataset = "https://huggingface.co/datasets/gnumanth/ninada-sanskrit-indic"
+
+[project.scripts]
+ninada = "ninada.cli:main"
+
+[project.optional-dependencies]
+neural = [
+  "onnxruntime>=1.18.0",
+  "numpy>=1.24.0",
+  "soundfile>=0.12.1",
+]
+train = [
+  "torch>=2.2.0",
+  "torchaudio>=2.2.0",
+  "transformers>=4.44.0",
+  "safetensors>=0.4.3",
+  "onnx>=1.16.0",
+  "onnxruntime>=1.18.0",
+  "huggingface_hub>=0.24.0",
+  "pyarrow>=15.0.0",
+  "scipy>=1.11.0",
+  "soundfile>=0.12.1",
+]
+dev = [
+  "pytest>=8.0.0",
+]
+
+[tool.setuptools.packages.find]
+where = ["python/src"]
+
+[tool.pytest.ini_options]
+pythonpath = ["python/src"]
+testpaths = ["tests"]

ninada-4.0.0/python/src/ninada/__init__.py

@@ -0,0 +1,76 @@
+"""ninada (निनाद): Streaming Pan-Indic & Classical Sanskrit Neural Text-to-Speech."""
+from .engine import (
+    HOP_LENGTH,
+    SAMPLE_RATE,
+    Ninada,
+    encode_wav_bytes,
+    speak,
+    split_prosodic_clauses,
+    stream,
+    synthesize_klatt_pcm,
+)
+from .g2p import (
+    BLANK_ID,
+    ID_TO_LANGUAGE,
+    ID_TO_TOKEN,
+    IPA_ID_TO_TOKEN,
+    IPA_VOCAB,
+    LANGUAGE_IDS,
+    RETROFLEX_LATERAL_ID,
+    TOKEN_TO_ID,
+    UNK_ID,
+    VOCAB_TOKENS,
+    WARMSTART_DONOR_MAP,
+    apply_phonological_rules,
+    brahmic_to_devanagari,
+    detect_script_and_lang,
+    iast_to_devanagari,
+    ipa_to_ids,
+    segment_aksharas,
+    text_to_ids,
+    text_to_ipa,
+)
+
+from .studio_g2p import (
+    STUDIO_HOP_LENGTH,
+    STUDIO_VOCABS,
+    clean_studio_text,
+    studio_text_to_ids,
+)
+
+__version__ = "4.0.0"
+
+__all__ = [
+    "__version__",
+    "SAMPLE_RATE",
+    "HOP_LENGTH",
+    "STUDIO_HOP_LENGTH",
+    "STUDIO_VOCABS",
+    "clean_studio_text",
+    "studio_text_to_ids",
+    "Ninada",
+    "speak",
+    "stream",
+    "split_prosodic_clauses",
+    "encode_wav_bytes",
+    "synthesize_klatt_pcm",
+    "VOCAB_TOKENS",
+    "TOKEN_TO_ID",
+    "ID_TO_TOKEN",
+    "IPA_VOCAB",
+    "IPA_ID_TO_TOKEN",
+    "RETROFLEX_LATERAL_ID",
+    "BLANK_ID",
+    "UNK_ID",
+    "LANGUAGE_IDS",
+    "ID_TO_LANGUAGE",
+    "WARMSTART_DONOR_MAP",
+    "detect_script_and_lang",
+    "iast_to_devanagari",
+    "brahmic_to_devanagari",
+    "apply_phonological_rules",
+    "segment_aksharas",
+    "text_to_ipa",
+    "ipa_to_ids",
+    "text_to_ids",
+]

ninada-4.0.0/python/src/ninada/cli.py

@@ -0,0 +1,43 @@
+"""Command-line interface for Ninada (Pan-Indic & Classical Sanskrit TTS)."""
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from pathlib import Path
+
+from .engine import SAMPLE_RATE, Ninada
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(
+        prog="ninada",
+        description="Ninada — Pan-Indic & Classical Sanskrit Neural TTS (24 kHz Studio WAV)",
+    )
+    parser.add_argument("text", nargs="?", help="Input text in Devanagari, Kannada, Tamil, Telugu, Malayalam, or IAST")
+    parser.add_argument("-l", "--lang", default=None, help="Language code: sa | hi | kn | ta | te | ml")
+    parser.add_argument("-o", "--output", type=Path, default=Path("ninada_out.wav"), help="Output WAV file path")
+    parser.add_argument("-r", "--rate", type=float, default=1.0, help="Speaking rate multiplier (default: 1.0)")
+    parser.add_argument("--tier", choices=["auto", "neural", "klatt"], default="auto", help="Synthesis engine tier")
+    parser.add_argument("--inspect", action="store_true", help="Print Pan-Indic G2P & akshara prosody JSON")
+    args = parser.parse_args()
+
+    txt = args.text if args.text else sys.stdin.read().strip()
+    if not txt:
+        parser.error("Provide input text as an argument or via stdin.")
+
+    engine = Ninada(lang=args.lang or "sa")
+    pcm, meta = engine.synthesize(txt, lang=args.lang, rate=args.rate, tier=args.tier)
+    if args.inspect:
+        print(json.dumps({k: v for k, v in meta.items() if k != "durations"}, ensure_ascii=False, indent=2))
+        return
+
+    from .engine import encode_wav_bytes
+
+    wav = encode_wav_bytes(pcm, sr=SAMPLE_RATE)
+    args.output.write_bytes(wav)
+    print(f"Wrote {len(wav)} bytes ({len(pcm)/float(SAMPLE_RATE):.2f}s, tier={meta['tier']}, lang={meta['lang']}) -> {args.output}")
+
+
+if __name__ == "__main__":
+    main()