Datasets:
| license: other | |
| task_categories: | |
| - automatic-speech-recognition | |
| language: | |
| - vi | |
| tags: | |
| - audio | |
| - vietnamese | |
| - gigaspeech2 | |
| pretty_name: GigaSpeech2 Vietnamese WAVs Missing Transcripts | |
| # GigaSpeech2 Vietnamese WAVs Missing Transcripts | |
| This dataset contains the Vietnamese GigaSpeech2 training WAV files whose IDs | |
| are absent from `train_refined.tsv`. | |
| ## Contents | |
| - 201,295 WAV files without a matching transcript | |
| - 41 uncompressed TAR shards in `shards/` | |
| - 36 GB of audio (approximately) | |
| - `processing_manifest.jsonl`: per-source-archive counts | |
| - `summary.json`: aggregate counts | |
| - `SHA256SUMS`: checksums for all TAR shards | |
| Each shard preserves the original relative layout: | |
| ```text | |
| release1/train/<archive>/<subfolder>/<utterance-id>.wav | |
| ``` | |
| The TAR files are intentionally uncompressed so they can be read sequentially | |
| and streamed without an additional decompression layer. | |