Datasets:
Search is not available for this dataset
audio audio |
|---|
End of preview. Expand in Data Studio
MiniMind PT-BR Kokoro TTS 68k
Synthetic Brazilian Portuguese speech dataset generated for the MiniMind/Tucano2 -> Mimi Talker fine-tuning experiments.
Contents
audio/: 68,486 WAV files generated with Kokoro TTS.manifests/kokoro_groq25k_audio_manifest.jsonl: source text manifest.manifests/kokoro_groq25k_audio_manifest_with_wavs.jsonl: manifest with WAV paths.manifests/kokoro_groq25k_audio_manifest_with_mimi.jsonl: manifest with Mimi token references.parquet/sft_t2a_kokoro_groq25k_ptbr.parquet: training parquet used by the MiniMind/Talker pipeline.
Intended Use
This dataset is meant for research and engineering experiments on Portuguese text-to-audio / audio-token fine-tuning:
- MiniMind-style Talker training.
- Tucano2 hidden-state to Mimi-token alignment.
- Brazilian Portuguese audio generation benchmarks.
Notes
The audio is synthetic and should be validated before use in production systems. The dataset was generated as part of an iterative fine-tuning workflow; manifests are included so the training format can be reconstructed.
Repository: marcosremar2/minimind-ptbr-kokoro-tts-68k
- Downloads last month
- 140