Human-transcribed & word-aligned

SilencioNetwork 's Collections

updated Sep 23

Current release standard: spontaneous speech, human-validated transcripts, word-level forced alignment. Built for evaluation, not training.

SilencioNetwork/swahili-speech

Viewer • Updated Sep 22 • 103 • 215

Note Kenyan Swahili. 103 speakers, one clip each — 66 Nairobi Sheng-influenced, 26 Mombasa coastal
SilencioNetwork/cebuano-speech

Viewer • Updated Sep 22 • 90 • 221

Note Long-form spontaneous Cebuano, 27,000+ timestamped tokens
SilencioNetwork/tagalog-filipino-speech

Viewer • Updated Sep 22 • 90 • 200 • 1

Note Spontaneous Tagalog/Filipino, 13,782 timestamped tokens
SilencioNetwork/yoruba-speech-transcribed

Viewer • Updated Sep 22 • 49 • 63

Note Spontaneous Yoruba, tone-marked. 30 speakers, 4,694 timestamped tokens
SilencioNetwork/hausa-speech-transcribed

Viewer • Updated Sep 22 • 49 • 49

Note Spontaneous Hausa, hooked consonants kept. 23 speakers, 2,912 timestamped tokens
SilencioNetwork/amharic-speech-transcribed

Viewer • Updated Sep 22 • 45 • 37

Note Spontaneous Amharic in Ge'ez script. 23 native speakers, 2,958 timestamped tokens