README.md

metadata

license: cc-by-4.0
language:
  - uk
task_categories:
  - automatic-speech-recognition
  - audio-classification
tags:
  - audio
  - speech
  - automatic-speech-recognition
  - speaker-diarization
  - datasets
pretty_name: LangPipe ASR Dataset
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files:
      - split: train
        path: train-*.parquet
dataset_info:
  features:
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: transcription
      dtype: string
    - name: transcription_original
      dtype: string
    - name: turns
      list:
        - name: speaker
          dtype: string
        - name: speaker_index
          dtype: int32
        - name: start
          dtype: float64
        - name: end
          dtype: float64
        - name: text
          dtype: string
    - name: has_pii
      dtype: bool
    - name: num_pii_objects
      dtype: int32
    - name: duration
      dtype: float64
    - name: speech_rms_dbfs
      dtype: float64
    - name: noise_rms_dbfs
      dtype: float64
    - name: snr_db
      dtype: float64
    - name: quality
      dtype: string
    - name: silence_ratio
      dtype: float64
    - name: speech_ratio
      dtype: float64
    - name: fillers
      list:
        - name: label
          dtype: string
        - name: language
          dtype: string
        - name: kind
          dtype: string
        - name: start
          dtype: float64
        - name: end
          dtype: float64
        - name: confidence
          dtype: float64
    - name: filler_count
      dtype: int32
    - name: obscene
      list:
        - name: term
          dtype: string
        - name: matched
          dtype: string
        - name: language
          dtype: string
        - name: start
          dtype: float64
        - name: end
          dtype: float64
        - name: confidence
          dtype: float64
    - name: obscene_count
      dtype: int32
    - name: code_switching
      dtype: bool
    - name: code_switching_type
      dtype: string
    - name: code_switching_confidence
      dtype: float64
    - name: code_switching_primary_language
      dtype: string
    - name: code_switching_secondary_language
      dtype: string
    - name: audio_language
      dtype: string
    - name: text_language
      dtype: string
    - name: speaker
      dtype: string
    - name: speaker_index
      dtype: int32
    - name: speaker_similarity
      dtype: float64
    - name: source_file
      dtype: string
    - name: loudness_lufs
      dtype: float64
    - name: loudness_range_lu
      dtype: float64
    - name: true_peak_dbtp
      dtype: float64
    - name: chunk_start
      dtype: float64
    - name: chunk_end
      dtype: float64

test-y

Yehor/test-y is an audio transcription dataset for automatic speech recognition and speech-processing experiments.

The dataset contains short audio clips together with text transcriptions and metadata such as duration, quality labels, signal/noise measurements, language information, filler annotations, code-switching indicators, and anonymized speaker labels.

Users can listen to the audio samples directly in the Hugging Face Dataset Viewer.

Dataset Structure

The dataset contains one split:

Split	Examples
train	1,000

Columns

• audio: audio sample.
• transcription: text transcription of the audio.
• duration: duration of the audio clip in seconds.
• speech_rms_dbfs: speech loudness estimate in dBFS.
• noise_rms_dbfs: noise loudness estimate in dBFS.
• snr_db: estimated signal-to-noise ratio.
• quality: quality label for the audio sample.
• silence_ratio: ratio of silence in the clip.
• speech_ratio: ratio of speech in the clip.
• fillers: detected filler words or discourse markers.
• filler_count: number of detected fillers.
• obscene: detected obscene terms, if any.
• obscene_count: number of detected obscene terms.
• code_switching: whether code-switching is detected.
• code_switching_type: type of detected code-switching, if available.
• code_switching_confidence: confidence score for code-switching detection.
• code_switching_primary_language: detected primary language.
• code_switching_secondary_language: detected secondary language, if available.
• audio_language: detected language of the audio.
• text_language: detected language of the transcription.
• speaker: anonymized speaker label.
• speaker_index: numeric speaker index.
• speaker_similarity: speaker similarity score.

Intended Uses

This dataset may be used for:

• automatic speech recognition;
• audio transcription research;
• speech quality analysis;
• multilingual and code-switching speech experiments;
• filler-word and speech-disfluency analysis.