diff --git a/docs/croissant-spec-draft.md b/docs/croissant-spec-draft.md
index 1e18661bc..4dab66ad5 100644
--- a/docs/croissant-spec-draft.md
+++ b/docs/croissant-spec-draft.md
@@ -1201,14 +1201,58 @@ Commonly used atomic data types:
| sc:Float |
Describes a float. |
+
+ | cr:Float16 |
+ Describes a float in half-precision floating-point format. |
+
+
+ | cr:Float32 |
+ Describes a float in single-precision floating-point format. |
+
+
+ | cr:Float64 |
+ Describes a float in double-precision floating-point format. |
+
| sc:Integer |
Describes an integer. |
+
+ | cr:Int8 |
+ Describes an 8-bit integer. |
+
+
+ | cr:Int16 |
+ Describes an 16-bit integer. |
+
+
+ | cr:Int32 |
+ Describes an 32-bit integer. |
+
+
+ | cr:Int64 |
+ Describes an 64-bit integer. |
+
| sc:Text |
Describes a string. |
+
+ | cr:UInt8 |
+ Describes an 8-bit unsigned integer. |
+
+
+ | cr:UInt16 |
+ Describes an 16-bit unsigned integer. |
+
+
+ | cr:UInt32 |
+ Describes an 32-bit unsigned integer. |
+
+
+ | cr:UInt64 |
+ Describes an 64-bit unsigned integer. |
+
Other data types commonly used in ML datasets:
@@ -1219,13 +1263,17 @@ Other data types commonly used in ML datasets:
Usage |
- | sc:ImageObject |
- Describes a field containing the content of an image (pixels). |
+ cr:AudioObject |
+ Represents a segment of audio as a digital sound recording. Refer to the section "ML-specific features > Bounding boxes". |
| cr:BoundingBox |
Describes the coordinates of a bounding box (4-number array). Refer to the section "ML-specific features > Bounding boxes". |
+
+ | sc:ImageObject |
+ Describes a field containing the content of an image (pixels). |
+
| cr:Split |
Describes a RecordSet used to divide data into multiple sets according to intended usage with regards to models. Refer to the section "ML-specific features > Splits". |
@@ -1789,6 +1837,25 @@ Bounding boxes are common annotations in computer vision. They describe imaginar
}
```
+
+### AudioObject
+
+Croissant uses Schema.org [AudioObject](https://schema.org/AudioObject) to represent an Audio feature. An AudioObject is a standard feature that represents a segment of audio as a digital sound recording. Croissant provides the audio-specific `cr:samplingRate` attribute, which can be specified at the audio field's `Source`:
+
+```json
+{
+ "@type": "cr:Field",
+ "@id": "recordset/audio",
+ "dataType": "sc:AudioObject",
+ "source": {
+ "fileSet": { "@id": "files" },
+ "extract": { "fileProperty": "content" },
+ "samplingRate": "16000",
+ }
+}
+```
+
+
### SegmentationMask
Segmentation masks are common annotations in computer vision. They describe pixel-perfect zones that outline objects or groups of objects in images or videos. Croissant defines `cr:SegmentationMask` with two manners to describe them: