[Addefan]

Azure TTS

The https://chatsemban.justdavi.dev/api/tts API endpoint stopped working (returning HTTP 500), which broke the Dmitry and Svetlana voices in the !tts command.

I deployed a compatible API endpoint with Azure voices on my own server and updated the endpoint URL in the code. I don't plan on taking it down in the foreseeable future, but in case it ever needs to be self-hosted elsewhere, here is the script to deploy the exact same endpoint:

# server.py

from fastapi import FastAPI, Query, Response
from fastapi.middleware.cors import CORSMiddleware
import edge_tts

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

@app.get("/api/tts")
async def tts(text: str = Query(...), voice: str = Query("ru-RU-DmitryNeural")):
    try:
        communicate = edge_tts.Communicate(text, voice)
        audio_data = bytearray()
        async for chunk in communicate.stream():
            if chunk["type"] == "audio":
                audio_data.extend(chunk["data"])
        return Response(content=bytes(audio_data), media_type="audio/mpeg")
    except Exception as e:
        return Response(content=str(e), status_code=500)

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)

Required dependencies: edge-tts, fastapi, and uvicorn (pip install edge-tts fastapi uvicorn).
Running python server.py starts the server at http://0.0.0.0:8000. HTTPS and a custom domain can be configured using NGINX + Let's Encrypt.

GIFs

All Twitch GIF messages have a multi-word fallback text format like [<title> GIF by <author>]. GIF rendering was already implemented in the project, but the unit tests only checked a single-word [GIF] string. Because the message text is tokenized by whitespace before rendering (createMessageTokenSnapshot), the full multi-word GIF substring stored in replacements never matched the individual word tokens during lookup and ended up being rendered as plain text.

Before tokenizing the message, I now replace the GIF substring with a whitespace-free placeholder so it stays a single token and properly matches the entry in replacements.

VoiceCatalog.tsx

Prior to any of these changes, tests/voiceCatalog.test.ts was failing in the original repository because it imported a .tsx file directly (outside Vite, Bun's built-in JSX transpiler attempts to import jsxDEV from solid-js, which does not exist at runtime).

I extracted the pure parsing logic from src/components/setup/VoiceCatalog.tsx into src/components/setup/voiceCatalogData.ts so the test imports a plain .ts module without JSX.

[Linaryx]

@Addefan Thanks 🤙