feat: add expressive offline Russian book TTS

tools/tts/prepare_sherpa_piper_model.py

@@ -0,0 +1,78 @@
+#!/usr/bin/env python3
+"""Prepare a Piper ONNX export for sherpa-onnx mobile inference."""
+
+import argparse
+import json
+import shutil
+from pathlib import Path
+
+import onnx
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", type=Path, required=True)
+    parser.add_argument("--config", type=Path, required=True)
+    parser.add_argument("--output-dir", type=Path, required=True)
+    args = parser.parse_args()
+
+    config = json.loads(args.config.read_text(encoding="utf-8"))
+    output_dir: Path = args.output_dir
+    output_dir.mkdir(parents=True, exist_ok=True)
+    output_model = output_dir / "aletheia_ru.onnx"
+    output_config = output_dir / "aletheia_ru.onnx.json"
+    output_tokens = output_dir / "tokens.txt"
+
+    shutil.copy2(args.model, output_model)
+    shutil.copy2(args.config, output_config)
+
+    id_map = config["phoneme_id_map"]
+    # sherpa-onnx's Piper lexicon maps one Unicode code point to one model ID.
+    # Newer Piper configs may also contain English diphthong aliases such as
+    # "aɪ". eSpeak emits their component code points, and sherpa rejects the
+    # multi-code-point aliases, so only the single-code-point table is mobile-safe.
+    token_rows = sorted(
+        ((ids[0], symbol) for symbol, ids in id_map.items() if len(symbol) == 1),
+        key=lambda row: row[0],
+    )
+    actual_ids = [row[0] for row in token_rows]
+    if len(actual_ids) != len(set(actual_ids)):
+        raise RuntimeError("phoneme_id_map contains duplicate token identifiers")
+    if not actual_ids or min(actual_ids) < 0 or max(actual_ids) >= config["num_symbols"]:
+        raise RuntimeError("phoneme_id_map contains a token outside the model symbol range")
+    output_tokens.write_text(
+        "".join(f"{symbol} {token_id}\n" for token_id, symbol in token_rows),
+        encoding="utf-8",
+        newline="\n",
+    )
+
+    model = onnx.load(str(output_model))
+    metadata = {
+        "model_type": "vits",
+        "comment": "piper",
+        "language": "Russian",
+        "voice": config["espeak"]["voice"],
+        "has_espeak": "1",
+        "n_speakers": str(config["num_speakers"]),
+        "sample_rate": str(config["audio"]["sample_rate"]),
+    }
+    existing = {item.key: item for item in model.metadata_props}
+    for key, value in metadata.items():
+        if key in existing:
+            existing[key].value = value
+        else:
+            item = model.metadata_props.add()
+            item.key = key
+            item.value = value
+    onnx.save(model, str(output_model))
+
+    print(json.dumps({
+        "model": str(output_model),
+        "model_bytes": output_model.stat().st_size,
+        "tokens": len(token_rows),
+        "metadata": metadata,
+    }, ensure_ascii=False))
+
+
+if __name__ == "__main__":
+    main()