---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: channel
    dtype: string
  - name: transcript_whisper
    dtype: string
  - name: title
    dtype: string
  - name: audio
    dtype:
      audio:
        sampling_rate: 16000
  - name: transcript_sensevoice
    dtype: string
  - name: emotion_sensevoice
    sequence: string
  - name: event_sensevoice
    sequence: string
  - name: c50
    dtype: float
  - name: snr
    dtype: float
  - name: speech_duration
    dtype: float
  - name: emotion_emotion2vec
    dtype: string
  splits:
  - name: train
    num_bytes: 544892035865.877
    num_examples: 1478373
  download_size: 527025543429
  dataset_size: 544892035865.877
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
task_categories:
- automatic-speech-recognition
- audio-classification
language:
- zh
- yue
---

## Cantonese Youtube Pseudo-Transcription Dataset
- Contains approximately 10k hours of audio sourced from YouTube
  - Videos are chosen at random, and scraped on a channel basis
  - Includes news, vlogs, entertainment, stories, health
- Columns
  - `transcript_whisper`: Transcribed using `Scrya/whisper-large-v2-cantonese` with `alvanlii/whisper-small-cantonese` for speculative decoding
  - `transcript_sensevoice`: Transcribed using `FunAudioLLM/SenseVoiceSmall`
    - used [OpenCC](https://github.com/BYVoid/OpenCC) to convert to traditional chinese
    - isolated event tags to `event_sensevoice`
    - isolated emotion tags to `emotion_sensevoice`
  - `snr`: Signal-to-noise ratio, extracted from `ylacombe/brouhaha-best`
  - `c50`: Speech clarity, extracted from `ylacombe/brouhaha-best`
  - `emotion`: Emotion, extracted from `emotion2vec/emotion2vec_plus_large`
  - Note that `id` does not reflect the ordering of the audio within the same video
- Processing
  - The full audio is split using [WhisperX](https://github.com/m-bain/whisperX), using `Scrya/whisper-large-v2-cantonese`
    - it is split in <30s chunks and according to speakers
  - Preliminary filtering includes filtering out phrases like:
    - `like/subscribe to YouTube channel`
    - `subtitles by [xxxx]`
    - Additional filtering is recommended for your own use
- Note: An earlier version of the dataset has duplicated data. I recommend re-downloading it if you downloaded it before Nov-7-2024