Multi-genre TV recordings from the British Broadcasting Corporation (BBC) covering a broad range of English-language broadcast output. The data includes audio and metadata for speech recognition, speaker diarization, and lightly supervised alignment tasks from the 2015 challenge.
Use Cases
- Train speech recognition models using the multi-genre TV audio and corresponding transcripts
- Develop speaker diarization algorithms to segment audio by individual speakers in broadcast settings
- Evaluate lightly supervised alignment techniques to synchronize text with the audio stream
Strengths
- Multi-genre TV recordings sourced from the British Broadcasting Corporation (BBC)
- Includes data for speech recognition, speaker diarization, and lightly supervised alignment
- Represents the full range of TV output from the 2015 MGB-1 Challenge