Configuration Parsing Warning:In adapter_config.json: "peft.task_type" must be a string

Daniel Rosehill: OmniVoice LoRA voice

The voice of Daniel Rosehill, the creator of the My Weird Prompts podcast, who appears on it in his own voice (https://myweirdprompts.com), baked into OmniVoice as a LoRA adapter, so it speaks in this voice without a reference clip.

Daniel Rosehill is a real person, and this adapter is trained on their own recordings, released by them: takes from Hebrish/Hebrish-TTS-Voice-Recordings (CC-BY-NC-4.0; read and spontaneous speech recorded for a voice clone). Published by them, for use as their voice on the podcast. Do not use it to impersonate them.

How it was made (2026-11-14)

  1. Takes: scripted takes of 2–20 s, digit-free, desktop-microphone takes first, then the rest, weighted toward expressive, storytelling, question and conversational sets for livelier prosody. 459 takes (40 min): conversational, narrative, expressive, personal, questions, long, news, code-switched and desktop-mic takes; Sidon kept for 231.
  2. Restore + polish: Sidon v0.1 speech restoration where it kept the words intact, else the raw take; a light "warm" EQ/compression chain and loudnorm to -19 LUFS, 24 kHz mono.
  3. Train: OmniVoice's LoRA recipe (source at git 08be0b4): r 16, alpha 32, lr 1e-4, 600 steps on one L40S. Published checkpoint: checkpoint-600.

Use

import torch
from omnivoice import OmniVoice
from omnivoice.utils.lora import load_lora_adapter  # in OmniVoice's git source; not in the 0.2.1 wheel

model = OmniVoice.from_pretrained("k2-fsa/OmniVoice", device_map="cuda:0", dtype=torch.float16)
model = load_lora_adapter(model, "<local snapshot of this repo>")
audio = model.generate(text="I've read the press release so you don't have to.", language="English",
                       num_step=32, guidance_scale=2.0, denoise=True, postprocess_output=True)[0]

Samples (no reference clip)

Text Audio
So here's this week's prompt, and it's one I've been chewing on for a while now, honestly. samples/prompt.mp3
I was standing in the queue at the makolet this morning, and it struck me that nobody actually knows how this works. samples/aside.mp3
So the people who built one of the best open voice cloners now build the tools that catch voice clones. samples/sentence.mp3

Licence

CC-BY-NC-4.0, following the OmniVoice weights it adapts (CC-BY-NC, trained on Emilia). Non-commercial use only.

Recipe and scripts: the My Weird Prompts monorepo, planning/research/tts-landscape-2026-11/pilot-5-herman-lora-2026-11-14/.

Downloads last month
30
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for My-Weird-Prompts/Daniel-OmniVoice-LoRA

Finetuned
Qwen/Qwen3-0.6B
Finetuned
k2-fsa/OmniVoice
Adapter
(60)
this model

Collections including My-Weird-Prompts/Daniel-OmniVoice-LoRA