Voice

Speak your task, review the text, and send it to EvoX.

Composer dictation turns ideas, task requirements, and follow-up instructions into editable text. Before sending, you can review the wording, attach files, or adjust the scope.

For existing recordings or generated narration, use the built-in speech plugin. Dictation and audio processing are separate workflows; real-time two-way voice calls are not available.

Start dictating

1. Open the conversation or task you want to continue.
2. Select the microphone in the composer and grant operating-system microphone access when prompted.
3. Say what you need, then finish dictation and review the text.
4. Correct names, numbers, and file names before sending.

Dictation prepares input text; finishing speech does not automatically execute the task. When continuing existing work, review the conversation context before adding new instructions.

Turn an idea into a task

Start with what you want done, then describe the materials, scope, and expected output. You can edit the text afterward rather than getting every word right on the first attempt.

For example:

• “Summarize this week's progress from these project notes, including completed work and open issues.”
• “Turn this introduction into a presentation opening script, keeping the key facts.”
• “Continue the previous task, but suggest changes before editing any files.”

Tasks retain the current conversation's permissions and approval steps. Voice input does not grant additional permission to modify files, share data, or publish.

Process recordings and generate speech

Use the built-in speech-generation plugin to work with an audio file or turn text into speech. It is enabled on first launch. If you turned it off, configure and re-enable it under Extensions > Installed.

Remote synthesis, cloning, and transcription require a saved active provider and its credentials. OpenAI, Google, and StepFun are supported. Remote calls through the plugin use the configured provider; local transcription needs no provider key.

• Transcribe recordings: Local whisper.cpp is the default, producing timestamped SRT subtitles or plain text when requested. Audio is uploaded only when a remote provider is explicitly selected.
• Generate narration: Browse voices, create a short preview, and then generate the complete audio.
• Reuse an authorized voice: Register a voice using reference audio and the consent evidence required by the provider.

Audio and transcripts are saved in their respective task artifact directories. The plugin does not accept arbitrary output directories. See Skills and plugins for plugin setup.

Prepare local transcription

Local transcription requires whisper-cli and a ggml model on your computer. EvoX bundles neither dependency. Put whisper-cli on the system PATH and the model in EvoX’s <agentDir>/models, or provide an absolute model path in the task. On macOS, install the executable with brew install whisper-cpp.

Test a short recording first and check names, timestamps, and the output file. If the executable or model is missing, follow the error to install it; do not switch to a remote service just to dismiss the error. When format conversion is needed, use WAV, MP3, FLAC, or OGG. Transcripts are saved under <agentDir>/artifacts/transcripts/ with new file names, leaving the source recording unchanged.

Dictation or audio transcription?

These capabilities do not include live voice conversion or real-time voice calls.

Permissions and recording safety

Dictation requires operating-system microphone access. Local plugin transcription does not upload recordings. Before selecting a remote provider, review the destination, recording upload scope, and provider billing. Per-call approval follows the current conversation’s approval tier. Full access automatically approves provider tool calls, so a confirmation card is not a guaranteed barrier against uploads.

Clone only your own voice or a voice you have explicit authorization to use. Required consent recordings must be genuinely spoken by the person concerned; EvoX cannot generate them or accept rights declarations on their behalf.

Troubleshooting

• The microphone does not respond: Check system permissions, your input device, and the microphone selected in EvoX.
• The text is inaccurate: Review names, numbers, and file paths, then edit before sending.
• Audio tools are missing: Check the speech-generation switch and connection in the installed list. For remote processing, also check the active provider and credentials.
• Local transcription cannot start: Check whisper-cli, the ggml model, and input format. Adding a remote key does not repair local transcription.
• You do not want recordings uploaded: Prefer local processing for audio-file transcription. Do not assume all dictation or remote services behave like local transcription.