myLocalTTS

A lightweight, offline speech-to-text and text-to-speech tool for Windows that runs locally using Whisper.cpp and Piper. Press a hotkey to record your voice, release to transcribe, and the text is automatically pasted into your active application. Or select text and press F10 to have it read aloud.

Features

• Fully Offline - No internet connection required, all processing happens locally
• GPU Accelerated - Supports CUDA for fast transcription on NVIDIA GPUs
• Auto Language Detection - Automatically detects the spoken language
• Speech-to-Text (F9) - Hold F9 to record, release to transcribe and paste
• Text-to-Speech (F10) - Select text and press F10 to read it aloud
• Microphone Selection - Choose your preferred input device at startup
• Low Latency - Text appears almost instantly after releasing the hotkey
• Cancellable TTS - Press F10 again while audio is playing to stop

Requirements

• Windows 10/11
• Whisper.cpp executable (whisper-cli.exe) - for Speech-to-Text
• A Whisper model file (e.g., ggml-large-v3-turbo.bin) - for Speech-to-Text
• Piper executable (piper.exe) - for Text-to-Speech (optional)
• A Piper voice model (.onnx file) - for Text-to-Speech (optional)
• For GPU acceleration: NVIDIA GPU with CUDA support

Installation

1. Download the release

Download the latest release from the Releases page.

2. Download Whisper.cpp

Note: The Whisper executable and DLLs are not included in this repository due to their size and licensing. You must download them separately.

1. Go to whisper.cpp releases

2. Download the appropriate version:

   • CPU only: whisper-bin-x64.zip
   • NVIDIA GPU (recommended): whisper-cublas-12.4.0-bin-x64.zip

   Tip: The whisper-cublas-12.4.0-bin-x64.zip version has been tested and works best for CUDA GPU acceleration.

3. Extract the ZIP file

4. Copy the following files to your myLocalTTS folder:

   • whisper-cli.exe (required)
   • All .dll files (required - these include ggml, CUDA, and other dependencies)

Example DLLs you should see (varies by version):

ggml.dll
whisper.dll
cublas64_*.dll      (CUDA version only)
cublasLt64_*.dll    (CUDA version only)
cudart64_*.dll      (CUDA version only)

3. Download a Whisper model

Note: Model files are not included in this repository due to their size (~75MB to ~3GB). You must download them separately.

Download a model from Hugging Face:

Model	Size	Quality	Speed
ggml-tiny.bin	~75 MB	Basic	Fastest
ggml-base.bin	~147 MB	Good	Fast
ggml-small.bin	~466 MB	Better	Medium
ggml-medium.bin	~1.5 GB	Great	Slower
ggml-large-v3-turbo.bin	~1.6 GB	Best	Fast (GPU)

For best results with a GPU, use ggml-large-v3-turbo.bin:

Invoke-WebRequest -Uri "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-v3-turbo.bin" -OutFile "ggml-large-v3-turbo.bin"

4. Download Piper (Optional - for Text-to-Speech)

Note: Piper is optional. If not configured, only Speech-to-Text (F9) will be available.

1. Go to Piper releases
2. Download piper_windows_amd64.zip
3. Extract the ZIP file - it contains a piper folder with all required files
4. Copy the entire piper folder to your myLocalTTS directory (or rename/move it as needed)

Important: The piper folder must contain all of these files for Piper to work:

piper/
├── piper.exe                         (main executable)
├── espeak-ng.dll                     (required DLL)
├── espeak-ng-data/                   (required folder - phoneme data)
│   └── ... (many files inside)
├── onnxruntime.dll                   (required DLL)
├── onnxruntime_providers_shared.dll  (required DLL)
├── piper_phonemize.dll               (required DLL)
└── libtashkeel_model.ort             (required model file)

Warning: If you only copy piper.exe without the DLLs and espeak-ng-data folder, you will get a "DLL not found" error (exit code -1073741515).

5. Download a Piper Voice Model (Optional)

Download a voice model from Piper Voices:

1. Choose a language folder (e.g., en/en_US for US English, pt/pt_BR for Brazilian Portuguese)
2. Choose a speaker (e.g., amy, lessac)
3. Choose a quality level (low, medium, or high)
4. Download both files:
   • The .onnx model file (the voice model itself)
   • The .onnx.json config file (model configuration - required)
5. Place both files in your piper folder

Example for US English (Amy voice, medium quality):

# Download to the piper folder
Invoke-WebRequest -Uri "https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/amy/medium/en_US-amy-medium.onnx" -OutFile "piper\piper-model.onnx"
Invoke-WebRequest -Uri "https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/amy/medium/en_US-amy-medium.onnx.json" -OutFile "piper\piper-model.onnx.json"

Example for Brazilian Portuguese (faber voice, medium quality):

Invoke-WebRequest -Uri "https://huggingface.co/rhasspy/piper-voices/resolve/main/pt/pt_BR/faber/medium/pt_BR-faber-medium.onnx" -OutFile "piper\piper-model.onnx"
Invoke-WebRequest -Uri "https://huggingface.co/rhasspy/piper-voices/resolve/main/pt/pt_BR/faber/medium/pt_BR-faber-medium.onnx.json" -OutFile "piper\piper-model.onnx.json"

Note: The .onnx.json file must have the same base name as the .onnx file. If you rename the model to piper-model.onnx, rename the config to piper-model.onnx.json.

6. Create the TTS Configuration File

Create a tts_config.txt file in your myLocalTTS folder to tell the application where to find Piper:

# Piper TTS Configuration
PIPER_PATH=piper\piper.exe
PIPER_MODEL=piper\piper-model.onnx

Tip: You can use relative paths (as shown above) or absolute paths.

7. Folder structure

Your folder should look like this:

myLocalTTS/
├── local_tts_tool.exe
├── whisper-cli.exe
├── ggml-large-v3-turbo.bin
├── tts_config.txt                    (required for TTS - points to piper)
├── piper/                            (optional - for TTS)
│   ├── piper.exe
│   ├── espeak-ng.dll
│   ├── espeak-ng-data/
│   ├── onnxruntime.dll
│   ├── onnxruntime_providers_shared.dll
│   ├── piper_phonemize.dll
│   ├── libtashkeel_model.ort
│   ├── piper-model.onnx              (voice model)
│   └── piper-model.onnx.json         (voice model config)
├── *.dll (from whisper.cpp release)

Usage

1. Run local_tts_tool.exe
2. Select your microphone from the list
3. Wait for the "Listening..." message

Speech-to-Text (F9)

4. Hold F9 to record your voice
5. Release F9 to transcribe and auto-paste the text

The transcribed text will be automatically pasted into whatever application is currently focused.

Text-to-Speech (F10)

4. Select text in any application (highlight it with your mouse or Shift+Arrow keys)
5. Press F10 to have the selected text read aloud
6. Press F10 again while audio is playing to stop playback

Building from Source

Prerequisites

• Rust (latest stable)
• Windows 10/11

Build

git clone https://github.com/marcelo-t-tomita/myLocalTTS.git
cd myLocalTTS
cargo build --release

The executable will be at target/release/local_tts_tool.exe.

Configuration

Text-to-Speech (Piper) Configuration

You can configure Piper paths using environment variables or a config file.

Option 1: Environment Variables

$env:PIPER_PATH = "C:\path\to\piper.exe"
$env:PIPER_MODEL = "C:\path\to\model.onnx"

Option 2: Config File

Create a tts_config.txt file in the same folder as the executable:

# Piper TTS Configuration
PIPER_PATH=C:\path\to\piper.exe
PIPER_MODEL=C:\path\to\model.onnx

Default Paths (if not configured):

• piper.exe in the application folder
• piper-model.onnx in the application folder

Changing the hotkey

Edit src/main.rs:

• F9 (Speech-to-Text): Line ~97
• F10 (Text-to-Speech): Line ~98

let is_f9_pressed = KeybdKey::F9Key.is_pressed();
let is_f10_pressed = KeybdKey::F10Key.is_pressed();

Available keys: F1Key through F12Key, or see inputbot documentation.

Changing the Whisper model

Edit src/main.rs line 47:

let path_to_model = "ggml-large-v3-turbo.bin";

Forcing a specific language

Edit src/transcribe.rs and add the language flag:

.arg("-l")
.arg("en") // or "pt", "es", "fr", etc.

PowerShell Alias (Optional)

Add this to your PowerShell profile (notepad $PROFILE):

function tts {
    Push-Location 'C:\path\to\myLocalTTS'
    .\local_tts_tool.exe
    Pop-Location
}

Then just type tts to start the tool.

Troubleshooting

"Model file not found"

Make sure the model file (e.g., ggml-large-v3-turbo.bin) is in the same folder as the executable.

"Whisper executable not found"

Make sure whisper-cli.exe is in the same folder as the executable.

"DLL not found" error

Copy all .dll files from the whisper.cpp release to the same folder as the executable.

Transcription is slow

• Use the CUDA version of whisper.cpp for GPU acceleration
• Use a smaller model (e.g., ggml-small.bin)

F9 key not working

• Make sure no other application is using F9 as a global hotkey
• Try running from a different terminal window

"TTS narrator not available" warning

This is normal if you haven't configured Piper. Speech-to-Text (F9) will still work. To enable Text-to-Speech:

1. Download Piper from releases - extract the entire folder, not just piper.exe
2. Download a voice model (.onnx) and its config file (.onnx.json) from Piper Voices
3. Create a tts_config.txt file pointing to the paths (see Installation section)

Piper fails with exit code -1073741515

This error means "DLL not found". Make sure you have all these files in your piper folder:

• espeak-ng.dll
• espeak-ng-data/ folder (with all its contents)
• onnxruntime.dll
• onnxruntime_providers_shared.dll
• piper_phonemize.dll

Piper fails with empty error message

Make sure:

• The .onnx.json config file exists alongside the .onnx model file
• Both files have the same base name (e.g., piper-model.onnx and piper-model.onnx.json)
• The espeak-ng-data folder is present

F10 says "No text selected"

Make sure you have text selected (highlighted) in the active application before pressing F10. The tool simulates Ctrl+C to copy the selection.

License

MIT License - see LICENSE for details.

Acknowledgments

• whisper.cpp - High-performance C/C++ implementation of OpenAI's Whisper
• OpenAI Whisper - Original speech recognition model
• Piper - Fast, local neural text-to-speech system