Turn PDF slides into a timed, spoken presentation.
AutoTalk is a Qt desktop application for preparing and presenting slide decks with AI-generated narration and local speech synthesis. Choose a PDF, a language and a target duration; add audience or conference context, select a voice, and review or present the resulting talk.
Narration is written through Codex app-server using your ChatGPT sign-in. Qwen3-TTS 1.7B generates speech locally. The interface is built with Python and PySide6 and follows the platform's appearance.
Status: development prototype, version 0.3. Linux is the primary tested platform. Windows and macOS implementations are available for development but still require native qualification. There is no stable release yet.
Linux editor with Volker Christian's DACHS 2026 presentation, “Kollaborative KI-gestützte Softwareentwicklung in der Lehre,” and English narration in Realtime mode.
- Tailor narration to an audience, conference website or written conference scope.
- Edit narration and generate or preview audio for individual slides or a whole talk.
- Use predefined voices, a recording of your own voice, or a designed voice.
- Prepare talks in ten languages and retain separate language versions.
- Present fullscreen with synchronized slide advancement, pause/resume and live demos.
- Export slides and narration as video or audio; record the desktop and system audio on Linux.
- Save projects, reuse matching generated audio and control when speech models occupy GPU memory.
| Mode | Workflow |
|---|---|
| Prepared | Review text and audio at your own pace. Prepare and start completes missing content before presenting. |
| Quick | Choose a PDF, language and duration, then Start. Uses the configured voice and model settings without conference setup. |
| Realtime | Start prepares an opening audio buffer, then continues preparing later slides during the presentation. |
Realtime is generated narration, not an audience conversation. Startup time includes any required model loading; slower synthesis can cause buffering. Target duration is approximate until audio has been generated and measured.
Presenter view with Chinese narration: current and next slides, timing and playback controls.
| Platform | Speech backend | Validation status |
|---|---|---|
| Linux x86_64 with NVIDIA GPU | vLLM-Omni streaming | Tested locally on an RTX 2000 Ada laptop with 8 GB VRAM |
| Windows 11 x64 with NVIDIA GPU | Qwen / PyTorch CUDA | Native testing pending |
| macOS 14+ on Apple Silicon | MLX Audio / Metal | Native testing pending |
Speech generation requires a supported GPU and compatible operating-system drivers. CPU/NPU inference and Intel Macs are outside the current prototype. Playback of already prepared audio does not require the speech model or a synthesis-capable GPU.
First-time speech setup downloads a private runtime and model files automatically. Internet access and several GB of downloads are required; allow approximately 40 GB of free space on Linux for the runtime, models and installation caches. Subsequent use reuses cached files.
Narration generation requires an eligible ChatGPT sign-in through Codex and is subject to account usage limits. Slide images, extracted text and supplied context are sent to Codex. Voice references and speech synthesis remain local.
Until a stable release is published, run from source or use a development artifact from a successful native build. GitHub artifact downloads require sign-in; these builds are not qualified releases.
For a source installation on Linux or macOS, use Python 3.12–3.14:
git clone https://github.com/SNodeC/AutoTalk.git
cd AutoTalk
python3 -m venv .venv
.venv/bin/python -m pip install -e .
.venv/bin/autotalkOn Windows, create the environment with py -3.12 -m venv .venv, then use
.venv\Scripts\python.exe -m pip install -e . and .venv\Scripts\autotalk.exe.
Linux Qt runtime packages and native packaging instructions are covered in the
build guide.
- Select Open PDF… and choose your slide deck.
- Set the duration and language; optionally choose a voice and conference context.
- Select a mode. Review individual slides in Prepared mode, or choose Start in Quick or Realtime.
- Enable recording before presenting if you want to save the live session. For desktop recording, use End presentation and save video to finish it.
Projects are saved in an autotalk subfolder beside the imported PDF. Use
Open saved talk… to return to a project.
- User guide: preparation, voices, settings, presentation and recording.
- Building and testing: source setup, native packages, Qt dependencies and architecture.
- Roadmap: planned work and remaining qualification.
- Verification record: dated engineering checks and limitations.
- Issue tracker: report problems with the OS, AutoTalk/Qt versions, reproduction steps and relevant operation logs.
Review wording, pronunciation, timing and recording behavior before using a talk at an event. Automated builds do not establish GPU compatibility on every machine.
AutoTalk's own code is licensed under the MIT License. Dependencies, model weights and third-party material retain their respective licenses; see third-party notices.