i4ds/spc_r_segmented

---
dataset_info:
features:
- name: id
dtype: string
- name: duration
dtype: float64
- name: audio
dtype: audio
- name: text
dtype: string
language:
- de
task_categories:
- automatic-speech-recognition
source_datasets:
- i4ds/spc_r
license: cc-by-4.0
---

# i4ds/spc_r_segmented

Diarized and segmented speech dataset derived from [i4ds/spc_r](https://huggingface.co/datasets/i4ds/spc_r).

## Description

Each row is a merged speech segment belonging to a single speaker. The source audio and SRT subtitles from `i4ds/spc_r` were processed with the following pipeline:

1. **Diarization** -- [pyannote/speaker-diarization-3.1](https://huggingface.co/pyannote/speaker-diarization-3.1) assigned speaker labels to each SRT segment based on temporal overlap.
2. **Merging** -- Consecutive SRT segments from the same speaker were merged when the silence gap between them was below a threshold (default 1.0s) and the resulting duration stayed within bounds (default 10--20s).
3. **Slicing** -- The merged time ranges were used to slice the original audio waveform. Each segment is encoded as FLAC.

## Columns

| Column | Type | Description |
|------------|---------|--------------------------------------------------|
| `id` | string | Unique identifier (`row{NNNNN}_seg{NNN}`) |
| `duration` | float64 | Segment duration in seconds |
| `audio` | audio | FLAC audio for the segment |
| `text` | string | Merged transcript text from the SRT segments |

## Usage

```python
from datasets import load_dataset

ds = load_dataset("i4ds/spc_r_segmented")
print(ds["train"][0])
```