AI :: aisuite_ollama_fastapi_moonshine_gemma_kokoto.sh

aisuite_ollama_fastapi_moonshine_gemma_kokoto.sh

#!/bin/bash

 ##
## This script creates a FastRTC server which is well suited for audio conversations.
##
## Credits: https://www.youtube.com/watch?v=NthuElhWDk0
###


function aisuite_check_python3() {
   python3 -V | grep 3.13 > /dev/null
}


function aisuite_install_ollama() {
    which ollama > /dev/null || curl -fsSL https://ollama.ai/install.sh | sudo sh
}


function aisuite_install_ollama_gemma3() {
    ollama list | grep gemma3:4b > /dev/null || ollama pull gemma3:4b
}


function aisuite_fastRTC_prepare() {
    mkdir -p ~/workspace/fastRTC

    cat << EOD > ~/workspace/fastRTC/pyproject.toml
[project]
name = "simple-fastrtc-voice-chat"
version = "0.1.0"
description = "A simple voice chat that can run on the cpu "
readme = "README.md"
requires-python = ">=3.13"

dependencies = [
"fastrtc[stt]>=0.0.19",
"kokoro-onnx>=0.4.7",
"loguru>=0.7.3",
"ollama>=0.4.7",
]
EOD

    cat <<EOD > ~/workspace/fastRTC/code.py
from fastrtc import ReplyOnPause, Stream, get_stt_model, get_tts_model
from ollama import chat

stt_model = get_stt_model(model="moonshine/base")
tts_model = get_tts_model(model="kokoro")

SYSTEM_PROMPT = "You are a Voice AI Assistant. Engage in a conversation with the user and answer their questions. Do not use emojis or formatting, like asterisks or faces."

def echo(audio):
    transcript = stt_model.stt(audio)
    response = chat(
        model="gemma3:4b",
        messages=[
            {"role": "system", "content": SYSTEM_PROMPT},
            {"role": "user", "content": transcript}
        ]
    )
    response_text = response['message']['content']
    for audio_chunk in tts_model.stream_tts_sync(response_text):
        yield audio_chunk

server_name = "0.0.0.0"
server_port = 7860
share_public_link = True
ssl_verify = False

opts = { 'server_name': server_name, 'server_port': server_port, 'share': share_public_link }
if not ssl_verify:
    opts = opts | { 'ssl_verify': False, 'ssl_keyfile': "key.pem", 'ssl_certfile': "cert.pem" }

stream = Stream(ReplyOnPause(echo), modality="audio", mode="send-receive")
stream.ui.launch(**opts)
EOD

    local HOSTNAME=$(hostname)
    openssl req -x509 -newkey rsa:4096 -keyout key.pem -out cert.pem -sha256 -days 365 -nodes -subj "/C=US/ST=New York/L=New York/O=My Organization/OU=My Department/CN=${HOSTNAME}"
}


function aisuite_fastRTC_build() {
    source ~/.virtualenvs/fastRTC/bin/activate
    pushd ~/workspace/fastRTC
    pip install .
    deactivate > /dev/null
    popd
}


function aisuite_fastRTC_prepare_and_build() {
    which ensurepip || sudo apt install -y python3-venv python3-pip
    [[ -d ~/.virtualenvs ]] || mkdir -p ~/.virtualenvs
    [[ -d ~/.virtualenvs/python313 ]] || python3 -m venv ~/.virtualenvs/fastRTC
    [[ -f ~/workspace/fastRTC/code.py ]] || aisuite_fastRTC_prepare
    [[ -f ~/workspace/fastRTC/simple_fastrtc_voice_chat.egg-info ]] || aisuite_fastRTC_build
}


function aisuite_fastRTC_run() {
    if [[ $(cat /proc/meminfo | grep MemTotal | sed 's/kB//' | sed 's/ //g' | cut -d: -f2) -gt 16384000 ]] ;then
        [[ -f ~/workspace/fastRTC/simple_fastrtc_voice_chat.egg-info ]] || aisuite_fastRTC_prepare_and_build
        source ~/.virtualenvs/fastRTC/bin/activate
        pushd ~/workspace/fastRTC
        python3 code.py # should use an .egg image instead
        deactivate > /dev/null
        popd
    else
        echo "FATAL: At least 16Gb of memory is required"; return 1
    fi
}

aisuite_check_python3 && aisuite_install_ollama && aisuite_install_ollama_gemma3 && aisuite_fastRTC_run