Can you provide .exe file

[mesouravcodes]

Please provide .exe file if possible.

[ServeurpersoCom]

Please provide .exe file if possible.

Hello, as with acestep.cpp, I'm putting the latest binary here:
https://www.serveurperso.com/temp/omnivoice.cpp-win64/

1. Reproduce the directory structure
2. Click on an example script, you get a .WAV :)

[ServeurpersoCom]

clone.cmd (Morgan Freeman sample for voice clone) :
clone.wav

tts.cmd (voice design) :
tts.wav

[mindplay-dk]

would be better to have a Dockerfile, I think? even just for buildcpu.sh, so we can test this quickly.

I spent my entire morning trying to get this build, but I'm absolute buffoon at C++

would be nice if we could just podman something and then boot right into it and try it? ☺️

[ServeurpersoCom]

I use Podman for many developments so I can be idempotent and not pollute anything. But for distributing portable binaries or compilers, it's not worth it; there are no dependencies. It works on a stock Debian system with build-essential and CUDA.

[mindplay-dk]

I didn't even know build-essential existed yesterday, no idea what to install or how 😅

I had to add make and python and other things that weren't on my OS.

FROM ubuntu:22.04

ENV DEBIAN_FRONTEND=noninteractive

# Install build dependencies (CMake, C++ compiler, BLAS)
RUN apt-get update && apt-get install -y --no-install-recommends \
    cmake \
    g++ \
    make \
    libopenblas-dev \
    git \
    curl \
    pkg-config \
    python3 \
    python3-pip \
    && rm -rf /var/lib/apt/lists/*

# Install Hugging Face CLI for downloading models
RUN pip3 install huggingface-hub --quiet

WORKDIR /app

# Copy local source (already cloned with submodules)
COPY . .

# Build CPU version
RUN ./buildcpu.sh

# Download models (CPU inference only, no CUDA/Vulkan needed)
# Note: fixed repo name (was incorrectly using omnivoice.cpp-GGUF instead of OmniVoice-GGUF)
RUN mkdir -p models && \
    hf download "Serveurperso/OmniVoice-GGUF" "omnivoice-base-Q8_0.gguf" --local-dir models && \
    hf download "Serveurperso/OmniVoice-GGUF" "omnivoice-tokenizer-F32.gguf" --local-dir models

# Default command: show help
CMD ["./build/omnivoice-tts", "--help"]

This is slop, I don't know what I'm doing - it compiles the CPU version only, I could not get the CUDA version to build.

I think you system devs tend to over estimate what "the rest of us" have installed or can figured out in an hour or two - I have essentially zero knowledge of any of the things you take for granted, just saying. Whether it comes as a Dockerfile or at least just the required apt commands, it would have been useful to me, if just for the right incantations to install the required tools... I am 35 years a Windows users, so I am fairly Linux illiterate. ☺️

[ServeurpersoCom]

No worries at all, and don't sell yourself short, your Dockerfile does the job for a CPU build. If you get stuck, just ask here: that's exactly what issues are for, and honestly I do this for fun, so helping out is part of the fun. Happy to walk you through the CUDA build or anything else whenever you need.

[mindplay-dk]

don't sell yourself short, your Dockerfile does the job for a CPU build

mostly to no credit of mine 😅

Happy to walk you through the CUDA build or anything else whenever you need.

wouldn't it make more sense to write down the steps in a Dockerfile? it isn't more or less work, is it? 😊

also the build times for CUDA seem to be incredibly long? I was on a laptop when I attempted it, and it almost melted... it might be better to build once and push an image? so that everyone can be up and running without asking you how? 🙈

also helps potentially deploy it in production. 🙂

[ServeurpersoCom]

I'm only using rootless Podman (Containerfile / Kubernetes manifest) on my end because I find Docker outdated (and over-complicated for rootless). But I haven't had time to do that on GitHub or to do CI/CD.

[mindplay-dk]

I'm only using rootless Podman (Containerfile / Kubernetes manifest) on my end because I find Docker outdated

oh, same! 100% 😄

I just keep using the word "docker" because Podman is just a different (better!) implementation of that 😊