Add TeraTTS remote-code interface

.gitattributes

@@ -1,35 +1,2 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
 *.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,78 @@
+---
+library_name: transformers
+pipeline_tag: text-to-speech
+tags:
+- onnxruntime
+- text-to-speech
+- russian
+- english
+- custom-code
+---
+
+# TeraTTSv2 ONNX
+
+TeraTTSv2 text-to-speech with ONNX Runtime, selectable teacher/distilled
+diffusion samplers, precomputed voice styles, CPU/CUDA execution, and streamed
+audio chunks.
+
+## Installation
+
+```bash
+pip install -r requirements.txt
+```
+
+`sounddevice` is only required for direct speaker playback. On Linux, install
+the system PortAudio library if it is not already present.
+
+## Load with Transformers
+
+```python
+from transformers import AutoModel
+
+tts = AutoModel.from_pretrained(
+    "TeraSpace/TeraTTSv2",
+    trust_remote_code=True,
+    provider="CPUExecutionProvider",
+    threads=6
+)
+
+waveform = tts.generate_speech(
+    "<en>Hello from TeraTTS.</en>",
+    voice="en_female",
+)
+```
+
+Russian `<ru>…</ru>` text is automatically stress-marked by the bundled
+RUAccent-derived ONNX runtime before inference. It makes no secondary model
+download and does not require the `ruaccent` package. Manual `+` markers
+remain unchanged:
+
+```python
+waveform = tts.generate_speech(
+    "<ru>На двери висит замок.</ru>",
+    voice="ru_female",
+)
+```
+
+Use `diffusion_model="teacher"` to load the adjustable-CFG teacher sampler.
+The default distilled sampler has CFG 3 baked into its weights.
+
+## Stream audio
+
+```python
+for chunk in tts.generate_speech_stream(
+    "<en>Streaming speech is ready.</en>",
+    voice="en_female",
+):
+    # Send float32 mono chunks (44,100 Hz) to a player or network client.
+    consume(chunk)
+```
+
+The remote code loads only the selected sampler graph plus shared ONNX graphs.
+For security, pin a specific Hub commit when using `trust_remote_code=True`.
+
+## Attribution
+
+The local Russian stress annotator and its assets are adapted from
+[RUAccent](https://github.com/Den4ikAI/ruaccent), Copyright 2026 Denis Petrov,
+under the MIT License. See `RUACCENT_NOTICE.txt`.

config.json

@@ -0,0 +1,10 @@
+{
+  "architectures": ["TeraTTSModel"],
+  "auto_map": {
+    "AutoConfig": "configuration_teratts.TeraTTSConfig",
+    "AutoModel": "modeling_teratts.TeraTTSModel"
+  },
+  "default_diffusion_model": "distilled",
+  "model_type": "teratts_onnx",
+  "sample_rate": 44100
+}

configuration_teratts.py

@@ -0,0 +1,17 @@
+"""Transformers configuration for the ONNX Runtime TeraTTS release."""
+
+from transformers import PretrainedConfig
+
+
+class TeraTTSConfig(PretrainedConfig):
+    model_type = "teratts_onnx"
+
+    def __init__(
+        self,
+        sample_rate: int = 44_100,
+        default_diffusion_model: str = "distilled",
+        **kwargs,
+    ) -> None:
+        super().__init__(**kwargs)
+        self.sample_rate = sample_rate
+        self.default_diffusion_model = default_diffusion_model

modeling_teratts.py

@@ -0,0 +1,120 @@
+"""Hugging Face remote-code entry point for the ONNX Runtime TeraTTS release."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Iterator
+
+import numpy as np
+from transformers import PreTrainedModel
+
+from .configuration_teratts import TeraTTSConfig
+# Keep the bundled stress-runtime module in Transformers' remote-code module
+# dependency set. ``teratts.py`` imports it at runtime after this entry point
+# has been loaded.
+from .teratts_ruaccent import RUAccent as _BundledRUAccent
+from .teratts import LoadedTTS, generate_speech, generate_speech_stream, load_model
+
+
+class TeraTTSModel(PreTrainedModel):
+    """A reusable ONNX Runtime speech generator loaded through ``AutoModel``."""
+
+    config_class = TeraTTSConfig
+
+    def __init__(self, config: TeraTTSConfig) -> None:
+        super().__init__(config)
+        self.runtime: LoadedTTS | None = None
+
+    @classmethod
+    def from_pretrained(
+        cls,
+        pretrained_model_name_or_path: str | Path,
+        *model_args: object,
+        config: TeraTTSConfig | None = None,
+        provider: str = "CPUExecutionProvider",
+        threads: int | None = None,
+        diffusion_model: str | None = None,
+        russian_stress: bool = True,
+        ruaccent_model_size: str = "turbo3.1",
+        ruaccent_device: str = "CPU",
+        **kwargs: object,
+    ) -> "TeraTTSModel":
+        """Download/load the release and initialize reusable ONNX sessions.
+
+        ``provider``, ``threads``, and ``diffusion_model`` are TeraTTS-specific
+        arguments. Standard Hub arguments such as ``revision``, ``token``,
+        ``cache_dir``, and ``local_files_only`` are forwarded to the snapshot
+        download when a Hub model ID is supplied.
+        """
+        if model_args:
+            raise TypeError("TeraTTSModel.from_pretrained accepts no positional model arguments")
+        source = Path(pretrained_model_name_or_path)
+        if source.is_dir():
+            release = source
+        else:
+            from huggingface_hub import snapshot_download
+
+            download_keys = {"revision", "token", "cache_dir", "local_files_only", "force_download"}
+            download_kwargs = {key: kwargs.pop(key) for key in list(kwargs) if key in download_keys}
+            if kwargs:
+                unexpected = ", ".join(sorted(kwargs))
+                raise TypeError(f"unexpected TeraTTS loading arguments: {unexpected}")
+            release = Path(
+                snapshot_download(repo_id=str(pretrained_model_name_or_path), **download_kwargs)
+            )
+        if config is None:
+            config = TeraTTSConfig.from_pretrained(release)
+        instance = cls(config)
+        instance.runtime = load_model(
+            release,
+            model=diffusion_model or config.default_diffusion_model,
+            provider=provider,
+            threads=threads,
+            russian_stress=russian_stress,
+            ruaccent_model_size=ruaccent_model_size,
+            ruaccent_device=ruaccent_device,
+        )
+        return instance
+
+    def _runtime(self) -> LoadedTTS:
+        if self.runtime is None:
+            raise RuntimeError("load this model with from_pretrained() before generating speech")
+        return self.runtime
+
+    def generate_speech(
+        self,
+        text: str,
+        voice: str,
+        *,
+        duration_scale: float = 1.0,
+        guidance: float = 3.0,
+        seed: int = 1234,
+    ) -> np.ndarray:
+        return generate_speech(
+            self._runtime(),
+            text,
+            voice,
+            duration_scale=duration_scale,
+            guidance=guidance,
+            seed=seed,
+        )
+
+    def generate_speech_stream(
+        self,
+        text: str,
+        voice: str,
+        *,
+        duration_scale: float = 1.0,
+        guidance: float = 3.0,
+        seed: int = 1234,
+        chunk_frames: int = 16,
+    ) -> Iterator[np.ndarray]:
+        yield from generate_speech_stream(
+            self._runtime(),
+            text,
+            voice,
+            duration_scale=duration_scale,
+            guidance=guidance,
+            seed=seed,
+            chunk_frames=chunk_frames,
+        )

requirements.txt

@@ -0,0 +1,6 @@
+numpy>=1.24
+onnxruntime>=1.16
+torch>=2.1
+transformers>=4.40
+huggingface_hub>=0.23
+sounddevice>=0.5