| | --- |
| | 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]) |
| | ``` |
| |
|
