Create README.md

README.md

@@ -0,0 +1,94 @@
+---
+license: apache-2.0
+task_categories:
+- text-to-speech
+- automatic-speech-recognition
+language:
+- ru
+size_categories:
+- 1M<n<10M
+pretty_name: Webinar Audio Dataset
+---
+
+# Webinar Audio Dataset
+
+## Dataset Description
+
+This dataset contains 850 hours processed webinar audio segments with corresponding metadata. Each audio file represents a segment extracted from webinar recordings, processed at 44.1kHz sample rate.
+
+### Dataset Summary
+
+- **Language**: Russian
+- **Task**: TTS, ASR, Quality Asessment
+- **Audio format**: MP3, 44.1kHz sample rate
+- **Structure**: Segmented audio files with JSON metadata
+
+## Dataset Structure
+
+### Data Fields
+
+#### Basic Information
+- `audio`: Audio data (44.1kHz sample rate, MP3 format)
+- `file_name`: Name of the audio segment file (format: `<original_name>_<idx>.mp3`)
+- `segment_index`: Index of the audio segment within the original webinar
+- `original_name`: Original name of the webinar recording
+
+#### Transcription and Timing
+- `text`: Transcribed text of the audio segment
+- `start`: Start time of the segment in seconds
+- `end`: End time of the segment in seconds
+- `words`: Word-level timestamps and confidence scores
+
+#### Speaker Information
+- `speaker`: Speaker identifier (e.g., "SPEAKER_00")
+
+#### Quality Metrics
+- `utmos`: It's not utmos, it's EMOS overall quality score
+- `emos_1`, `emos_2`, `emos_3`: EMOS quality scores
+- `noise_confidence`: Noise detection confidence
+
+![design](https://huggingface.co/datasets/ESpeech/ESpeech-webinars2/resolve/main/mos.png)
+
+
+#### Segment Structure
+- `segment_type`: Type of segment ("merged" or "individual")
+- `num_sentences`: Number of sentences (for merged segments)
+- `original_segments`: Original subsegments data (for merged segments)
+
+#### VAD (Voice Activity Detection)
+- `vad_trimmed`: Whether VAD trimming was applied
+- `vad_start`: VAD start time
+- `trim_ratio`: Ratio of trimmed audio
+
+### Data Splits
+
+- **Train**: All available webinar segments
+
+## Dataset Creation
+
+### Source Data
+
+The dataset consists of webinar recordings that have been processed and segmented. Each webinar is split into multiple audio segments, with each segment saved as a separate MP3 file.
+
+## Usage
+
+### Loading the Dataset
+
+Load all zip files and unpack using ```zip -F webinars.zip --out webinars_fixed.zip && unzip webinars_fixed.zip```
+
+## Additional Information
+
+### Licensing Information
+
+MIT License - see LICENSE file for details.
+
+### Citation Information
+
+```bibtex
+@dataset{webinar_audio_dataset,
+  title={Webinar Audio Dataset},
+  authors={Denis Petrov}
+  year={2025},
+  url={https://huggingface.co/datasets/ESpeech/ESpeech-webinars2/}
+}
+```