Upload seamless_communication/datasets/datatypes.py with huggingface_hub
Browse files
seamless_communication/datasets/datatypes.py
ADDED
|
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Copyright (c) Meta Platforms, Inc. and affiliates
|
| 2 |
+
# All rights reserved.
|
| 3 |
+
#
|
| 4 |
+
# This source code is licensed under the license found in the
|
| 5 |
+
# MIT_LICENSE file in the root directory of this source tree.
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
from dataclasses import dataclass
|
| 9 |
+
from typing import Any, Dict, List, Optional
|
| 10 |
+
|
| 11 |
+
import torch
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
@dataclass
|
| 15 |
+
class MultimodalSample:
|
| 16 |
+
id: int
|
| 17 |
+
lang: str
|
| 18 |
+
text: str
|
| 19 |
+
audio_local_path: Optional[str] = None
|
| 20 |
+
waveform: Optional[torch.Tensor] = None
|
| 21 |
+
sampling_rate: Optional[int] = None
|
| 22 |
+
units: Optional[List[int]] = None
|
| 23 |
+
|
| 24 |
+
@classmethod
|
| 25 |
+
def from_json(cls, js: Dict[str, Any]) -> "MultimodalSample":
|
| 26 |
+
return cls(
|
| 27 |
+
id=js["id"],
|
| 28 |
+
lang=js["lang"],
|
| 29 |
+
text=js["text"],
|
| 30 |
+
audio_local_path=js.get("audio_local_path"),
|
| 31 |
+
waveform=None, # don't serialize
|
| 32 |
+
sampling_rate=js.get("sampling_rate"),
|
| 33 |
+
units=js.get("units"),
|
| 34 |
+
)
|
| 35 |
+
|
| 36 |
+
|
| 37 |
+
@dataclass
|
| 38 |
+
class LangPairSample:
|
| 39 |
+
source: MultimodalSample
|
| 40 |
+
target: MultimodalSample
|
| 41 |
+
|
| 42 |
+
@classmethod
|
| 43 |
+
def from_json(cls, js: Dict[str, Any]) -> "LangPairSample":
|
| 44 |
+
return cls(
|
| 45 |
+
source=MultimodalSample.from_json(js["source"]),
|
| 46 |
+
target=MultimodalSample.from_json(js["target"]),
|
| 47 |
+
)
|