feat(onnx): добавлен каталог моделей GigaAM

- Зачем:
  - добавлена локальная транскрипция смешанной речи и русского текста с пунктуацией.
- Что:
  - onnx-asr обновлён до 0.12 и зарегистрированы три модели GigaAM.
  - выбор compute_type учитывает опубликованные квантизации и явность настройки.
  - обновлены тесты, README и требования PRD.
- Проверка:
  - uv run pytest -q: 223 passed, 1 skipped.
  - выполнены smoke- и полные прогоны четырёх GigaAM-моделей.

src/local_transcriber/backends/__init__.py

@@ -32,7 +32,7 @@ def get_backend(device: str, *, compute_type_explicit: bool = True) -> Backend:
             raise ValueError(
                 "onnx-asr бэкенд недоступен. Установите: pip install onnx-asr[cpu,hub]"
             ) from None
-        return OnnxAsrBackend()
+        return OnnxAsrBackend(compute_type_explicit=compute_type_explicit)
 
     # cuda, cpu и всё остальное → faster-whisper
     from .faster_whisper import FasterWhisperBackend

src/local_transcriber/backends/onnx_asr.py

@@ -3,14 +3,46 @@
 from __future__ import annotations
 
 from collections.abc import Callable
+from dataclasses import dataclass
 from pathlib import Path
 from typing import Any
 
 from local_transcriber.types import Segment, TranscribeResult
 
+
+@dataclass(frozen=True)
+class OnnxModelSpec:
+    """Имя onnx-asr и опубликованные варианты квантизации модели."""
+
+    model_id: str
+    quantizations: frozenset[str | None]
+
+
+_INT8_AND_FLOAT32 = frozenset({"int8", None})
+
+MODEL_CATALOG: dict[str, OnnxModelSpec] = {
+    "gigaam-v3": OnnxModelSpec("gigaam-v3-ctc", _INT8_AND_FLOAT32),
+    "parakeet-v3": OnnxModelSpec(
+        "nemo-parakeet-tdt-0.6b-v3",
+        _INT8_AND_FLOAT32,
+    ),
+    "gigaam-multilingual-ctc": OnnxModelSpec(
+        "gigaam-multilingual-ctc",
+        _INT8_AND_FLOAT32,
+    ),
+    "gigaam-v3-e2e-ctc": OnnxModelSpec(
+        "gigaam-v3-e2e-ctc",
+        _INT8_AND_FLOAT32,
+    ),
+    "gigaam-v3-e2e-rnnt": OnnxModelSpec(
+        "gigaam-v3-e2e-rnnt",
+        _INT8_AND_FLOAT32,
+    ),
+}
+
+# Оставлено как совместимое представление публичного каталога алиасов.
 MODEL_ALIASES: dict[str, str] = {
-    "gigaam-v3": "gigaam-v3-ctc",
-    "parakeet-v3": "nemo-parakeet-tdt-0.6b-v3",
+    alias: spec.model_id for alias, spec in MODEL_CATALOG.items()
 }
 
 SUPPORTED_ALIASES = ", ".join(MODEL_ALIASES)
@@ -45,7 +77,8 @@ def _normalize_quantization(compute_type: str) -> str | None:
 class OnnxAsrBackend:
     """Бэкенд транскрипции через onnx-asr (ONNX Runtime)."""
 
-    def __init__(self):
+    def __init__(self, compute_type_explicit: bool = True):
+        self._compute_type_explicit = compute_type_explicit
         self.actual_compute_type: str | None = None
         self._resolved_model_id: str | None = None
         self._vad: Any = None
@@ -61,7 +94,26 @@ class OnnxAsrBackend:
         onnx-asr downloads models automatically via load_model(),
         so this just validates the alias and returns the identifier string.
         """
-        self.actual_compute_type = compute_type
+        spec = MODEL_CATALOG.get(model_name)
+        quantization = _normalize_quantization(compute_type)
+        if spec is not None and quantization not in spec.quantizations:
+            if self._compute_type_explicit:
+                available = _format_compute_types(spec.quantizations)
+                raise ValueError(
+                    f"Модель '{model_name}' недоступна с compute_type='{compute_type}' "
+                    f"для onnx-asr. Доступные варианты: {available}"
+                )
+
+            resolved_compute_type = _preferred_compute_type(spec.quantizations)
+            _notify(
+                on_status,
+                f"Модель {model_name} недоступна с compute_type={compute_type}; "
+                f"использую {resolved_compute_type}.",
+            )
+        else:
+            resolved_compute_type = _compute_type_for_quantization(quantization)
+
+        self.actual_compute_type = resolved_compute_type
         self._resolved_model_id = self._resolve_model(model_name)
         return self._resolved_model_id
 
@@ -79,7 +131,8 @@ class OnnxAsrBackend:
         """
         import onnx_asr
 
-        quantization = _normalize_quantization(compute_type)
+        actual_compute_type = self.actual_compute_type or compute_type
+        quantization = _normalize_quantization(actual_compute_type)
 
         model = onnx_asr.load_model(
             model=model_path,
@@ -113,7 +166,9 @@ class OnnxAsrBackend:
         segments: list[Segment] = []
         detected_language = language or "unknown"
 
-        for vad_seg in model.recognize(audio_array, sample_rate=16000, language=language):
+        for vad_seg in model.recognize(
+            audio_array, sample_rate=16000, language=language
+        ):
             seg = Segment(
                 start=max(0.0, vad_seg.start),
                 end=max(0.0, vad_seg.end),
@@ -149,6 +204,22 @@ class OnnxAsrBackend:
         )
 
 
+def _format_compute_types(quantizations: frozenset[str | None]) -> str:
+    values = [_compute_type_for_quantization(value) for value in quantizations]
+    return ", ".join(sorted(values))
+
+
+def _compute_type_for_quantization(quantization: str | None) -> str:
+    return "float32" if quantization is None else quantization
+
+
+def _preferred_compute_type(quantizations: frozenset[str | None]) -> str:
+    for quantization in ("int8", None, "fp16"):
+        if quantization in quantizations:
+            return _compute_type_for_quantization(quantization)
+    raise ValueError("Для ONNX-модели не указаны доступные квантизации")
+
+
 def _notify(on_status: Callable[[str], None] | None, message: str) -> None:
     if on_status is not None:
         on_status(message)

src/local_transcriber/config.py

@@ -1,10 +1,9 @@
 """Загрузка конфигурации из ``.transcriber.toml`` и каскад приоритетов."""
 
+import tomllib
 import warnings
 from pathlib import Path
 
-import tomllib
-
 HARDCODED_DEFAULTS: dict[str, str] = {
     "model": "medium",
     "language": "ru",
@@ -23,7 +22,15 @@ DEVICE_DEFAULTS: dict[str, dict[str, str]] = {
 
 # Одно место правды для допустимых ключей конфига
 _VALID_KEYS = set(HARDCODED_DEFAULTS)
-_VALID_DEVICES = {"auto", "cpu", "cuda", "openvino", "openvino-gpu", "openvino-cpu", "onnx"}
+_VALID_DEVICES = {
+    "auto",
+    "cpu",
+    "cuda",
+    "openvino",
+    "openvino-gpu",
+    "openvino-cpu",
+    "onnx",
+}
 
 
 def find_config_file() -> Path | None: