Posts starting with // are personal and outside of the parody, not always serious.
I just wanted my triple shot eXpresso, pineapple, licorice and goat milk latte.
I just wanted my triple shot eXpresso, pineapple, licorice and goat milk latte.
I'll need to look into why this is so slow. Maybe it is because I don't use streaming, maybe it's the hardware or the Omnivoice model.
I'll need to look into why this is so slow. Maybe it is because I don't use streaming, maybe it's the hardware or the Omnivoice model.
In terms of transcribing speech-to-text Hviske is very fast, I hardly notice any delay.
I just did a quick test in audio.cpp directly on a short command.
For 3353ms recording, it took 468ms to transcribe or ~14%
In terms of transcribing speech-to-text Hviske is very fast, I hardly notice any delay.
I just did a quick test in audio.cpp directly on a short command.
For 3353ms recording, it took 468ms to transcribe or ~14%
*Whispering* Also it's funnier this way *Pushes a 5 liter bucket of popcorn in to your hands*
*Whispering* Also it's funnier this way *Pushes a 5 liter bucket of popcorn in to your hands*
You can use any model audio.cpp supports, but I use these:
ASR: Hviske-v5.3 (Danish ASR model)
TTS: Omnivoice (instruct, not clone, voice cloning is fucking gross).
LLM: mradermacher/munin-qwen3.5-9B-i1-GGUF (Fine-tuned for Danish)
You can use any model audio.cpp supports, but I use these:
ASR: Hviske-v5.3 (Danish ASR model)
TTS: Omnivoice (instruct, not clone, voice cloning is fucking gross).
LLM: mradermacher/munin-qwen3.5-9B-i1-GGUF (Fine-tuned for Danish)
So please stop yelling or at least wait till I've gotten my pumpkin spice Raktajino.
So please stop yelling or at least wait till I've gotten my pumpkin spice Raktajino.
G'kar: So do I.
G'kar: So do I.