Practical overview
When it fits
Use it when you want local audio inference across several task types and can provide the relevant model files and task inputs. Its documented CLI examples are useful when you need explicit control over task, model, backend and file paths.
Who it is for
- People who want to run supported audio-model tasks locally through a WebUI.
- Developers using a native C++ CLI for speech, voice or music inference.
- Users seeking a shared runtime for several audio task types without a Python dependency.
Documented uses
- Generate speech from text with a local model.
- Transcribe an audio file using an ASR model.
- Convert a voice using an audio input and reference recording.
- Explore diarization, voice activity detection and music-generation tasks through the application.
Requirements & limits
- The core runtime is C++ and uses ggml; the project states that it has no Python dependency.
- The CLI exposes task, model, backend and offline or streaming mode selectors, but streaming availability depends on the model.
- The README describes Windows, Linux and macOS support; the CLI lists CPU, CUDA, Vulkan and Metal backend choices.
- Model files and inputs are local paths in the documented CLI examples, so each task depends on having appropriate assets available.
- The README labels JSON pipeline support experimental.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Generate speech with the documented PocketTTS CLI example
Use the CLI with a local PocketTTS model and text input to request a WAV output.
Before you start
- The audio.cpp CLI binary.
- A local PocketTTS model at the path supplied to --model.
- A CUDA backend for the documented example.
- Text to synthesize and the referenced sample voice file.
Steps
Select the tts task and pocket_tts family.
Set the model path and CUDA backend.
Provide text and the voice reference at assets/resources/sample.wav.
Set an output path such as build/out/pocket_tts.wav.
Expected outcome
The example specifies a WAV output path for the requested speech generation.
Guide limitations
- The example requires local model and voice-reference paths; it does not document model acquisition or installation.
- The CLI example is a documented recipe, not a verified execution result.
Transcribe audio with the documented Qwen3 ASR CLI example
Pass an audio file to the CLI for automatic speech recognition.
Before you start
- The audio.cpp CLI binary.
- A local Qwen3 ASR model at the path supplied to --model.
- A CUDA backend for the documented example.
- An audio file such as assets/resources/sample_16k.wav.
Steps
Select the asr task and qwen3_asr family.
Set the model path and CUDA backend.
Pass the audio file using the --audio option.
Expected outcome
The CLI is configured to run an ASR request on the specified audio input.
Guide limitations
- The example does not specify a transcript output file or model-download procedure.
- The CLI example is a documented recipe, not a verified execution result.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- kigner/audio.cpp-webui · Official homepage title · retained in its source language