Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
audio
audioduration (s)
4.03
6

Vessel Local voices

Short reference clips (4-6 s, 24 kHz mono) for the private Local voice of Hermes Vessel. The Local voice (MiniCPM-o 4.5) speaks in the voice and accent of whichever clip it is given (zero-shot cloning); the app lists each one so people can listen to a preview and pick. Vessel's installer downloads this repository at a pinned revision.

All clips are of English speakers who agreed to their voices being recorded and released for speech research, or are synthetic. None is a celebrity, a character or a commercial assistant voice.

id Name in the app Accent Kind Source
na_6319 June North American real LibriTTS-R speaker 6319
na_7850 Nora North American real LibriTTS-R speaker 7850
na_84 Grace North American real LibriTTS-R speaker 84
af_heart Heart North American synthetic Kokoro-82M voice af_heart
af_bella Bella North American synthetic Kokoro-82M voice af_bella
uk_p225 Alice British (Southern England) real VCTK speaker p225
uk_p257 Harriet British (Southern England) real VCTK speaker p257
uk_p268 Sophie British (Southern England) real VCTK speaker p268
bf_emma Emma British synthetic Kokoro-82M voice bf_emma
bf_isabella Isabella British synthetic Kokoro-82M voice bf_isabella

The names are labels chosen for the app; the real speakers are anonymous in their corpora. voices.json holds the same information for software.

Sources, licenses and changes

  • VCTK clips (uk_*): from the CSTR VCTK Corpus: English Multi-speaker Corpus for CSTR Voice Cloning Toolkit (version 0.92), Junichi Yamagishi, Christophe Veaux and Kirsten MacDonald, University of Edinburgh, The Centre for Speech Technology Research. Licensed CC BY 4.0. Taken from the mic1 recordings via the Hugging Face mirror sanchit-gandhi/vctk.
  • LibriTTS-R clips (na_*): from LibriTTS-R (Y. Koizumi et al., 2023), a restored version of LibriTTS (H. Zen et al., 2019), itself derived from LibriVox public-domain audiobook recordings. Licensed CC BY 4.0. Taken from the dev.clean split via mythicinfinity/libritts_r; speakers chosen with the Parler-TTS speaker descriptions (parler-tts/libritts-r-filtered-speaker-descriptions).
  • Kokoro clips (af_*, bf_*): generated with Kokoro-82M by hexgrad (Apache-2.0), reading one passage written for this repository.

Changes made to the CC BY 4.0 recordings: several sentences by the same speaker were joined with short pauses, silence was trimmed, the result was cut to at most 6 seconds at a pause and peak-normalised, and it was resampled to 24 kHz mono. The clips are offered under the same licenses as their sources; please keep this attribution when reusing them.

Use

Speak in one of these voices only as the voice of your own assistant; don't use them to imitate or pass as a real person.

Downloads last month
49