Text to Speech & Recognition

Manage and upgrade your Android voices
Onyx TTS is an Android voice manager, speech recognition tool, and system TTS bridge for:
• Local Piper ONNX voices
• Custom HTTP voice servers
• ElevenLabs voices using your own API key
Onyx makes these voices available to compatible apps through Android’s standard text-to-speech framework.
Upgrade a Piper voice with ElevenLabs
For voices you own or have permission to use, Onyx can create an ElevenLabs counterpart from an installed Piper voice.
Onyx generates about 20 seconds of Piper speech for you to preview. If approved, it uploads the sample to ElevenLabs and creates a voice with the same name so both versions pair automatically.
When paired:
• ElevenLabs provides the expressive online voice
• Piper remains available as an offline backup
• Onyx returns to Piper if ElevenLabs fails or responds too slowly
Results vary. Cloning synthetic speech may produce lower quality than clean original recordings. An ElevenLabs API key, eligible account, and permission to use the voice are required.
Local, cloud, and self-hosted voices
Piper ONNX
Install compatible models for fast, private speech generated on your device without an internet connection.
ElevenLabs
Use existing ElevenLabs voices or create a counterpart from an eligible Piper voice.
Custom HTTP voices
Connect compatible local or online TTS endpoints, including models running on your computer, home server, or cloud server.
Speech recognition
Onyx includes a local Whisper model for private, offline transcription. You can also use ElevenLabs speech recognition with automatic local fallback.
Each Tasker Speech to Text action can:
• Follow the default selected in Onyx
• Use ElevenLabs first with local fallback
• Use local Whisper only
Tasker automation
Onyx provides Tasker plug-in actions for:
• Say
• Say to File
• Speech to Text
Tasker can select Piper, ElevenLabs, or HTTP voices and speak literal text or Tasker variables. The variable tag button displays variables available to the current task and variables previously entered into Onyx.
Say to File creates a standard WAV in shared Downloads storage. For example, Tasker/tts_output.wav is saved as Download/Tasker/tts_output.wav.
Speech to Text accepts compatible audio files and stores the transcription in your selected Tasker variable.
Actions can control network access and whether Tasker waits for completion. Say can also select the Android audio stream and audio-focus behavior.
Use your voices across Android
After setting Onyx as your system TTS engine, compatible accessibility tools, screen readers, ebook readers, document readers, communication tools, and productivity apps can use your voices.
Google Maps does not use Onyx for navigation. Apps with their own speech systems may also ignore Android’s selected TTS engine.
Features
• Manage Piper, ElevenLabs, and HTTP voices
• Clone eligible Piper voices into ElevenLabs counterparts, with ElevenLabs as the main voice and Piper as the local fallback.
• Preview reference audio before uploading
• Pair ElevenLabs voices with Piper fallbacks
• Connect local or remote HTTP voice servers
• Use voices through Android system TTS
• Transcribe locally with Whisper
• Use ElevenLabs STT with local fallback
• Add speech, WAV generation, and transcription to Tasker
• Keep the built-in Lessac fallback for reliable basic speech
Privacy and control
Piper speech and local Whisper transcription run on your device.
ElevenLabs receives text sent for speech and audio sent for online transcription. Voice creation uploads the Piper reference only after you approve it.
HTTP voices receive text at the server address you configure.
Onyx lets you balance expressiveness, privacy, reliability, and performance.

Recent changes:
Default language - en-US
• Added language support for compatible English, German, Spanish and Ukrainian Piper voices.
• Content-aware pronunciation smoothing handles English words such as read and lead, German context-sensitive words, Spanish numbers and units, and Ukrainian word stress.
• Imported Piper voices retain their language, region and sample rate, and incompatible models are refused.
• English is included; other language voices are imported separately.