czaky / dreame_voice_pack

Voice Packs for Dreame Vacuum Robots

Uses voice generation by Uberduck.AI.

Works at least with L10s Ultra, L10 Pro, Z10 Pro and W10. Optimized for valetudo use.

Current voice packs:

• GLaDOS voice pack - hash code
• Tiff voice pack - hash code
• Sweetie voice pack - hash code
• Fluttershy voice pack - hash code

Installation

1. In Valetudo go to "Robot Settings" -> "Misc Settings"

2. Enter the following information in the "Voice packs" section:

   • URL: https://github.com/czaky/dreame_voice_pack/raw/master/[character]/voice.tar.gz
     • for example: https://github.com/czaky/dreame_voice_pack/raw/master/glados/voice.tar.gz
   • Language Code: CUSTOM - or anything but language codes like EN
     • for example: GLADOS
   • Hash: Lookup in the [character]/HASH.txt file
     • for example: bc10b417ace1dd1aaf4f3e1525aa3b73

3. Click "Set Voice Pack"

4. Check if the *.ogg files have been copied to /data/personalized_voice/CUSTOM/ directory on the robot.

5. If not, extract the voice.tar.gz and copy the *.ogg files using your favorite scp tool.

Generation

Prerequisites

1. Install ffmpeg and vorbis-tools

sudo apt install ffmpeg vorbis-tools

2. Install uberduck python API

pip install uberduck

Invocation

1. Sing up with https://uberduck.ai
2. Generate a public key and a secret token in your profile.
3. Replace the public keys and tokens in the keys.csv
4. Run the script (replacing kirby-tiff with one of the voices)

python generate.py --voice kirby-tiff [--volume 2] [--normalize]

The default voice is glados.

Note: Normalization uses ffmpeg and may create random artifacts in the resulting audio. It is not recommended with Uberduck.AI voice characters.

Note: voice list can be found at https://app.uberduck.ai/quack-help . Not all voices are good. Please, consult: https://app.uberduck.ai/leaderboard/voice

Listen to the resulting .ogg files for any artifacts. Change the text if necessary.

Modify

The defaults are loaded first and are found in the defaults folder. Then the values in each character folder override the defaults. Each folder can have following files:

• ./sayings.csv - a list of saying IDs and saying texts. If the text is empty, generator will look for a prepared .ogg file.
• ./replacement.csv - a list of words and their replacements, in case the AI characters cannot pronounce those correctly.
• ./*.ogg - prepared ogg audio files, used when the text for the saying ID is empty.

Results

In each voice-character folder:

• ./voice.tar.gz - archive of the voices generated.
• ./HASH.txt - contains the md5 hash code of the voice archive.
• ./ogg/*.ogg - resulting ogg audio files, that go into the archive.
• ./tts/*.wav - transformed text audios with name being the hash of the text.

Contribute

Please, don't add any obscenity or racist stuff. We cannot accept this on a public directory. There is no point in offending anybody. Reserve this stuff for your local installation.

Also, please, don't add anything copy-protected.

Thanks

Thanks to https://github.com/ccoors/dreame_voice_packs and https://github.com/Findus23/voice_pack_dreame for the inspiration and https://uberduck.ai for the voice generation.