npub1kpxeczgxu5xld0m5af4eqzhzjuw7vd7z650w2tph4kkh3vek032qsnuak4 (npub1kpx…uak4) numen is quite cool, I didn't know of it before! But it comes with a couple of cons after some testing:
1. It takes forever to compile.
2. It seems to be intended as a stand-alone executable that transcribes the speech, while my use-case is more that of a programmable voice assistant flow (with hotword detection, speech detection that goes on only on hotword, and programmable events that I can hook to). Maybe it's worth looking a bit into the vosk API.
3. At least the small-en-us model seems to struggle a bit with my non-US accent (and apparently even with that of a native speaker), but maybe it can be solved by just using a larger model I guess.
Fabio Manganiello /
npub13u…dpvgs
2024-02-07 22:54:37
in reply to nevent1q…uttx
Author Public Key
npub13uunvh7djw9ep54nswkuxlneyee7ehcpc7e53t68krykrdeg6j4qrdpvgsPublished at
2024-02-07 22:54:37Event JSON
{
"id": "af192f9bcbf7c902425cec1873c2016481ef04469d5058c337daf1d5a30e6839",
"pubkey": "8f39365fcd938b90d2b383adc37e792673ecdf01c7b348af47b0c961b728d4aa",
"created_at": 1707346477,
"kind": 1,
"tags": [
[
"p",
"b04d9c0906e50df6bf74ea6b900ae2971de637c2d51ee52c37adad78b3367c54",
"wss://relay.mostr.pub"
],
[
"p",
"7360361a0ef8fb82562ec3eeeb758c0017bc36eba77ac773ffe6e4a45b3f6bc1",
"wss://relay.mostr.pub"
],
[
"e",
"0b6ebb8cdc3438f91291be365aef0190044ec9bcc92ce3ea3cd62d7b66f002f5",
"wss://relay.mostr.pub",
"reply"
],
[
"proxy",
"https://manganiello.social/objects/8b1f64ac-7801-4cd8-b4bb-499d22ee0a14",
"activitypub"
]
],
"content": "nostr:npub1kpxeczgxu5xld0m5af4eqzhzjuw7vd7z650w2tph4kkh3vek032qsnuak4 numen is quite cool, I didn't know of it before! But it comes with a couple of cons after some testing:\n\n1. It takes forever to compile.\n\n2. It seems to be intended as a stand-alone executable that transcribes the speech, while my use-case is more that of a programmable voice assistant flow (with hotword detection, speech detection that goes on only on hotword, and programmable events that I can hook to). Maybe it's worth looking a bit into the vosk API.\n\n3. At least the small-en-us model seems to struggle a bit with my non-US accent (and apparently even with that of a native speaker), but maybe it can be solved by just using a larger model I guess.",
"sig": "b0e0e6af1766bd8dafbfac75876f617f5d75281fa2d3fa07c5c17d77f18d776fa7ef14e06ead99b8aaac4192feea23afcc9c597a421dc07281a1aa794294225e"
}