benow-conversation

A .NET text-to-speech tool and OpenAI-compatible proxy that converts text and LLM responses to spoken audio via the OpenRouter API.

Technology

• .NET 10 (ASP.NET Core for proxy mode)
• OpenRouter API — TTS synthesis using multiple providers (OpenAI, etc.)
• Serilog — structured logging to console and file
• ffmpeg / ffplay — audio format conversion and playback

Prerequisites

• .NET 10 SDK
• ffmpeg (optional — enables PCM conversion and audio playback via ffplay)
• An OpenRouter API key

Configuration

Copy appsettings.json and create appsettings.Development.json (git-ignored) with your API key:

{
  "OpenRouter": {
    "ApiKey": "sk-or-..."
  }
}

Set the environment variable for development:

export DOTNET_ENVIRONMENT=Development

How to Run

# Restore and build
dotnet build src/benow-conversation

# Run from the project directory
dotnet run --project src/benow-conversation -- "Hello, world"

CLI Mode

Pass text directly or via a file. Audio is saved to output/ or streamed to speakers.

# Basic synthesis (streams to speakers if no --output)
dotnet run --project src/benow-conversation -- "Hello, world"

# Save to file
dotnet run --project src/benow-conversation -- "Hello, world" --output hello.mp3

# Use a persona
dotnet run --project src/benow-conversation -- "Hello" --persona female-1

# Stream audio in real-time
dotnet run --project src/benow-conversation -- "Hello" --stream

# List available voices or models
dotnet run --project src/benow-conversation -- --list-voices
dotnet run --project src/benow-conversation -- --list-models

Daemon / Proxy Mode

Runs an OpenAI-compatible HTTP proxy. Chat completion responses are automatically converted to speech.

dotnet run --project src/benow-conversation -- --daemon

The proxy listens on 0.0.0.0:8080 by default. Point any OpenAI-compatible client at http://localhost:8080/v1 and chat responses will be spoken aloud. The proxy forwards requests to OpenRouter, extracts the response text, and queues it for TTS playback.

Key CLI Options

Flag	Description
--persona <name>	Use a named persona (model, voice, instructions)
--voice <id>	Override voice ID (all for all voices)
--model <id>	Override TTS model (all for all providers)
--output <path>	Save audio to file or directory
--stream	Stream audio to speakers in real-time
--play / --no-play	Control playback behavior
--daemon	Run as proxy with automatic TTS
--temperature <n>	Sampling temperature (0–2)
--seed <n>	Seed for reproducible output
--save-persona <name>	Save current settings as a named persona
--openai-instructions	Style instructions for OpenAI models

How It Works

CLI mode takes text input, sends a TTS request to OpenRouter's /audio/speech endpoint, and either saves the resulting audio to a file or streams it to speakers via ffplay. If a model doesn't support the requested format, it falls back to PCM and converts via ffmpeg.

Daemon mode starts a lightweight ASP.NET Core web server that acts as an OpenAI-compatible proxy. It intercepts /v1/chat/completions requests, forwards them to OpenRouter, streams the response back to the client, and simultaneously extracts the text content for TTS. A background SpeechQueue processes texts sequentially — if a new response arrives while one is being spoken, the current speech is cancelled and the new one begins.

Personas store a complete TTS configuration (model, voice, instructions, temperature, seed) under a name, allowing quick switching between voices and styles. Output profiles store device and volume preferences for audio playback.

Project Structure

src/benow-conversation/
  Program.cs               Entry point, CLI parsing, DI setup
  appsettings.json          Base configuration (no secrets)
  Configuration/
    AppSettings.cs          Settings models
  Models/
    TtsModelInfo.cs         TTS model metadata
    TtsRequest.cs           TTS API request models
  Services/
    AudioConverter.cs       PCM-to-MP3 conversion via ffmpeg
    AudioPlayer.cs          Audio playback via ffplay
    ITtsService.cs          TTS service interface
    TtsService.cs           Core TTS synthesis logic
    ModelService.cs         OpenRouter model/voice discovery
    ProxyService.cs         OpenAI-compatible proxy server
    SpeechQueue.cs          Queued TTS playback for daemon mode
tests/benow-conversation.Tests/
  Unit and integration tests

Tests

dotnet test tests/benow-conversation.Tests