Setup guide

Set up what is missing, then return to your task

Configuration links in Voice Studio, interpretation and audio import open the relevant setup step. A task strip names where you will return. Choose Save & return from that step without repeating all five steps.

1. Leave your text, speech script, reference transcript or import file in the task, then open its configuration link.
2. Update the required model or service and choose Save & return. The interface stays loaded, applies the saved changes and returns to the same workspace or dialog.
3. Choose Return without saving, close, or press Esc to discard only the unsaved setup changes. Your original task draft remains.
4. Check the model, reference and languages before starting again yourself. Returning does not start recording, generation, uploads or model downloads.

A failed save keeps the form for retry. If settings were saved but applying or returning did not finish, choose Retry return; it will not save twice. Audio import retains the file path, method and language. First-run setup still finishes with the save action on its last step.

1 · Microphone

Choose the device that captures your speech and check system microphone permission and input level. Remote meeting audio and your microphone may be different sources. Hearing someone does not prove the recording input captures them. Verify the input with a short recording first.

12

1. 1Select the input device you use.
2. 2Record a short sample and listen to check the input.

2 · Hotkey

RightAlt starts or stops recording by default. You can change it here. Try it in the main window, then in Floating Transcription. If it does not respond, check that the app is running and that another app or input method is not using the key. Keep only one app instance open.

12

1. 1Check the current recording shortcut.
2. 2Click here and press your preferred shortcut.

3 · Models & Params

Register each local model once, then select it by name in a workspace. Download the files first and open Setup guide → Local model library.

1. Choose its use and family.Use Speech recognition for transcription and Speech synthesis for cloning or design. Translation, speaker recognition and emotion recognition have their own uses.
2. Name each variant clearly.Register Qwen Base · 0.6B and Qwen VoiceDesign · 1.7B separately. Enter the actual size or quantization, such as FP16 or Q4_K_M, in the optional field.
3. Browse and register.Check Choose a folder for a model directory; uncheck it for a single ONNX or GGUF file. Select your own storage location, click Register model, then finish setup to save.

After registration, choose its name in Model used for transcription at the top of the library, check its files, then expand Try transcribing a recording. To get files, use Prepare download beside the selected family or expand Download models and components below the library. Review the source and destination disk before explicitly starting a download. Registration changes the draft; finish and save Setup to keep it.

Register each size or quantization separately, even within the same family. FP16 and Q4_K_M below illustrate registration; running a model requires a complete package compatible with the app backend. Entering a variant name does not convert the model. For VoxCPM2, set VoxCPM2 runtime folder (optional) in the same entry if a separate runtime is needed. Voice Studio and interpretation use that saved runtime when you choose the entry; no second path is required. Leave it empty to use the default runtime. This does not start a download.

12

1. 1Keep the original folder paired with its actual variant and a clear name.
2. 2Register another variant separately. Check, edit or remove this individual entry.

Select in a workspace without entering the path again

In Voice Studio, open Model Setup → Local models and choose a Registered voice model by name. Transcription, interpretation, voiceprints, emotion and local translation use their own registered-model lists. Interpretation can use a different recognition model from the main transcription view.

12

1. 1Select the saved model by name and variant. Select the reference recording separately.
2. 2Use Manage local models to add a variant or change a path in setup.

12

1. 1Choose Local native model, then a registered speech model. Recognition and speech selections are saved separately.
2. 2Return to the library if the model is missing. Finish setup to save, then select it here.

Use Manage local models to return to the library. After editing, finish setup to save, then return to the workspace and select the entry. Closing setup discards unsaved changes.

Check and prepare

Check files inspects the package; it does not prove successful inference. For cloning, also select a reference recording, use Check model, and audition a short sentence. Check again after changing the reference.

If conversion is required, use Prepare native model. Preparation adds and selects a Native entry while keeping the original entry and directory. The Index package is added at .boundless-flow/native inside the selected directory, without moving it to C:. That disk still needs write permission and free space. Remove entry does not delete model files.

12

1. 1The original entry and directory stay available. You can inspect, edit or remove its registration.
2. 2Preparation adds a Native entry pointing to the derived folder. Select it, check and audition.

4 · API Config

Search by service or capability, expand a card and enter Base URL, API Key and one actual available model ID. An example ID does not install a model. For translation, select the configured service in Translation service and use Start translation test with a short passage. Finish setup to save. In Interpretation Workbench, choose that saved Translation service instead of entering its endpoint, key and model again.

The translation test shows the current source and target languages. Its source comes from Models & languages. Test again after changing languages. Default examples can follow a known source language; your edited text is kept, so check that it matches the setting. A response that clearly conflicts with the target language is discarded without an automatic retry. Check the settings and text before trying again manually. Language checks do not replace reviewing the translation.

Saving configuration does not validate an API. LLM, ASR, TTS and realtime capabilities can require different authentication and protocols. A chat-compatible URL may not support speech or bidirectional streaming. In this guide, configure and audition the matching streaming adapter. Configure cloning model-server URLs and keys under Models & Params. Workspaces select the saved connection; languages, reference audio, voices, emotion and playback remain task-specific choices.

12

1. 1Search for a service or capability.
2. 2Expand the service you want to configure.

1234

1. 1Select a configured translation service.
2. 2Enter a short passage to test.
3. 3Choose the target language.
4. 4Run the test after entering valid settings.

12

1. 1Select NVIDIA GPU, or keep CPU.
2. 2Leave the CUDA field empty for automatic discovery, or choose a file.

Check what you need for this task

In the final setup step, choose a task under What would you like to do?, then click Check current settings. The default Recording & subtitles view checks the application, input device and recognition model. It does not require voice cloning or a notes text model.

1. Choose LingLu notes, Local voice cloning or Segment interpreting when preparing that task. Interpreting checks the active speech route: model service, speech API, local native model or compatibility synthesis.
2. If an item needs attention, use the Next button below the results. It opens the existing setup step; you do not enter the model path in a second place.
3. After basic checks, use Next: test transcription or the task's audition action. These buttons open the test area; they do not start recording or call a service automatically.
4. Review all saved settings is folded initially. Expand it when needed, then use Complete and save to keep your changes. Closing setup still discards the unsaved draft.

File checks and configured API parameters do not establish recognition quality, microphone permission or paid service access. A short trial is still needed. Generating notes or translation sends text to the selected service; a model-service voice route also receives reference audio. The task description explains what applies.

If settings change, run checks again. Copy redacted summary excludes keys, addresses, paths and trial content. A LAN Ollama notes connection can remain keyless.

5 · Usage and saving

Complete the last step on first use, or choose Save & return when setting up an existing task. Reopening setup starts from saved configuration. Model and service drafts are retained separately within the current setup flow; closing without finishing discards the changes.

Settings backups lets you inspect candidates and explicitly restore one. Check its source and scope first. Damaged configuration is not treated as a successful read; existing files are retained before recovery.

1

1. 1On first use, finish setup to save. For an existing task, choose Save & return. Closing discards unsaved setup changes.

Choose once, use in each workspace

Register each local model in Setup with a recognizable name and its size or quantization in the variant field. Save separate profiles for different variants. VoxCPM runtime folders are saved with the model.

Workspaces show the model’s purpose, name and saved variant. Older entries named after a backend, such as ONNX, show their folder name to help tell them apart. A displayed name does not verify the files or their capabilities; use Check model before a short trial.

In the translation studio, speech recognition converts input to text; segment speech voices a completed Mini recording; continuous cloned speech voices translated segments with the selected reference. Select the models for the mode you intend to use. Manage models opens Setup; there is no need to enter the path again.

For supported continuous parameters, turn the dial or focus it and use arrow keys. The number field remains available for precise entry. During operation, unavailable controls are dimmed and cannot change the running model.