EdgeTX - Voice packs

This repository contains the files needed to generate the voice packages used in EdgeTX.

The currently supported languages are:

• Chinese Mandarin
• Chinese Taiwan Mandarin
• Chinese Hongkong Cantonese
• Chilean Spanish
• Czech
• Danish
• English
• French
• German
• Italian
• Japanese
• Korean
• Polish
• Portuguese
• Russian
• Spanish
• Swedish
• Ukrainian

The following languages are not yet supported:

• Dutch
• Hungarian
• Slovak

Directory structure

SOUNDS

This folder has the audio files already processed and separated by language.

To use them, the language folder (for example, en) must be under the SOUNDS folder of your SDCARD. With the folder added, go to the EdgeTX settings menu and select the language of the audio language that will be used (eg English).

To use any audio on your switches, first copy the file you want to use to your language folder, then you can use this file in your Global Functions or Special Functions by selecting a switch for the function and choosing the Play track option.

SCRIPTS

Inside the language folder there is a folder called SCRIPTS, which has audio files for commonly used LUA scripts. These audio files are generated with the same voice as the other audio files of their language pack. Each script has their own folder.

BETAFLIGHT

Audio files for Betaflight TX Lua Scripts. Copy the WAV files from SOUNDS/<lang>/SCRIPTS/BETAFLIGHT/ to SOUNDS/en/ to overwrite the original audio files of the script.

INAV

Audio files for iNav Lua Telemetry Flight Status. Copy the WAV files from SOUNDS/<lang>/SCRIPTS/INAV/ to SCRIPTS/TELEMETRY/iNav/<lang>/ to overwrite the original audio files of the script.

YAAPU

Audio files for Yaapu Telemetry Script and Widget. Copy the WAV files from SOUNDS/<lang>/SCRIPTS/YAAPU/ to SOUNDS/yaapu0/<lang>/ to overwrite the original audio files of the script.

Korean (ko-KR)

The Korean voice pack provides full support for native Korean speakers using EdgeTX.

• Voice files were generated using Google Cloud Text-to-Speech (Wavenet-B).
• Generation activity can be traced via the public Google Cloud Logs Console.
• Format: 16-bit PCM WAV, Mono, 32kHz – fully compatible with EdgeTX audio requirements.
• Files are located in: SOUNDS/ko
• Indexed and mapped using:
  • voices/ko-KR.csv — main system phrases
  • voices/ko-KR_scripts.csv — numeric and special script terms
• This voice pack was personally developed and contributed by @siyeongjang, a Korean RC enthusiast, to improve accessibility and user experience for Korean-speaking pilots.
• Feedback or improvement suggestions are welcome. Please feel free to open an issue or leave a comment in the Pull Request.

Polish (pl-PL)

Files are generated using ElevenLabs voice synthesis. There are 10k tokens available for free.

Voice: Sarah

ElevenMultilingual v2.

One file at a time

Ubuntu commands to prepare wav file from mp3:

• Normal way:

  ffmpeg -i ElevenLabs_2025-09-03T16_24_21_Sarah_pre_sp100_s50_sb75_se0_b_m2.mp3 -ar 32000 -ac 1 -sample_fmt s16 engstp.wav

• To cut out words added for correct accent (0.5sec from beginning):

  ffmpeg -ss 0.5 -i ElevenLabs_2025-09-03T16_24_21_Sarah_pre_sp100_s50_sb75_se0_b_m2.mp3 -ar 32000 -ac 1 -sample_fmt s16 engstp.wav

• Many files at once from the same folder:

  for f in *.mp3; do
  ffmpeg -i "$f" -ar 32000 -ac 1 -sample_fmt s16 "${f%.mp3}.wav"
  done

Automated way

sudo apt update
sudo apt install libportaudio2 portaudio19-dev libasound-dev -y
pip install elevenlabs python-dotenv

If you don't have venv configured or enabled:

python3 -m venv .venv
source .venv/bin/activate

Set API key environment variable:

export ELEVENLABS_API_KEY="<API_KEY>"

Optionally add your languages to array:

languages = [
    ("voices/pl-PL.csv", "EXAVITQu4vr4xnSDxMaL", "pl"),
    # Add other languages here
]

And finally run script to generate all the files. Script is smart and skips already generated files to not use tokens from ElevenLabs.

python3 voice-gen-elevenlabs.py

Voices

All of the voices used in the EdgeTX voice packs have been picked from the neural voices offered by Microsoft Azure text to speech service, in order to get as close as possible to human-like voices. If you want to see what voices are available, and try different phrases, check out the online demo generator. Using some recording software, you could even save your own phrases and use them in the voice packs.

Generating custom phrases

If you have a Azure Speech Services subscription (there is a free usage tier), phrases can be generated with curl or a http client like postman. After building a text to speech resource in Azure you can use it by REST calls (http requests).

The request url is: https://<YOUR_RESOURCE_REGION>.tts.speech.microsoft.com/cognitiveservices/v1

You should add the following headers to your request:

Ocp-Apim-Subscription-Key: <YOUR_RESOURCE_KEY>
Content-Type: application/ssml+xml
X-Microsoft-OutputFormat: riff-8khz-16bit-mono-pcm

Note: EdgeTX supports up to 32khz .wav file but in that range 8khz is the highest value supported by the conversion service. However, it is possible to select higher quality like riff-48khz-16bit-mono-pcm and convert to 32khz afterwards with another tool (i.e. ffmpeg -i input.wav -ar 32000 output.wav) if you want the best possible audio quality.

And in the request body (raw) place your ssml (change the voice name according to your preference, the full list is here):

<speak version='1.0' xml:lang='en-US'>
    <voice xml:lang='en-US' xml:gender='Female' name='en-US-MichelleNeural'>YOUR_PHRASE_HERE</voice>
</speak>

In order to tweak some parameters of voice generation (i.e. rate, pitch) refer to the SSML markup documentation for values and the generate.sh script to see how they are applied.

How to build yourself

In order to generate the voice packages and do the release processing, you will need a Linux environment to run in. Ubuntu 18.04 is recommended as it is a LTS release. Newer versions and other flavours of Linux will most likely work also, but are not supported.

You will also need to have ffmpeg, spx and ffmpeg-normalize packages installed.

ffmpeg is used to clip any silence from the audio files. ffmpeg-normalise is used to normalise the audio files. spx is the tool that generates the audio files using Microsoft Azure Text to Speech processing.

Installing SPX can be a little tricky, but can be installed as follows:

wget https://packages.microsoft.com/config/ubuntu/20.04/packages-microsoft-prod.deb -O packages-microsoft-prod.deb
sudo dpkg -i packages-microsoft-prod.deb
sudo apt-get update; \
  sudo apt-get install -y apt-transport-https && \
  sudo apt-get update && \
  sudo apt-get install -y dotnet-sdk-6.0

dotnet tool install --global Microsoft.CognitiveServices.Speech.CLI

After you have installed SPX, you will also need to create a Microsoft Azure account if you don't have one already. There are both free and paid options, but the free one is sufficient for this purpose - it is just rate limited. After you have done that, follow the quick start guide to configure the required region and subscription keys.

Alternatives

• Mike has created a python script that can be used to generate the audio using Googles Text to Speech service - https://github.com/xsnoopy/edgetx-sdcard-sounds
• The OpenTX Speaker voice generator (Windows only) uses the built in text to speech engine of Microsoft Windows, and can be used to generate new audio also. https://www.open-tx.org/2014/03/15/opentx-speaker
• Record your operating systems own text-to-speech narration capability
• Get audio from Google Translate - as shown in #106
• Google Docs - using text to speech addons
• TTSMAKER
• NaturalReader
• Speechify
• ReadSpeaker
• iSpeech
• Amazon Polly
• macOS say command (with Siri voices) then add -ar 16000 -ac 1 -sample_fmt s16 to ffmpeg

Contributing

See CONTRIBUTING.md