Boundless FlowDesktop voice workspace · User guide
0.5.0简体中文
DOUBAO SPEECH · REALTIME ASR

Doubao Speech · Realtime transcription

See text while speaking, with sentences segmented and confirmed by Doubao Speech. Save the connection once in Setup, then select it when recording.

Save the connection before recording

Save the Doubao Speech connection and enabled recognition resource in Setup, then select the service in Recording settings. Return to Setup when changing the resource; do not enter the key again on other pages. First try a short recording with your own account and input device to check access, recognition quality and charges. Speed and quality depend on the service, connection and recording conditions.

Before you start

  • Enable the required Doubao Speech recognition service/resource and obtain its current speech-console API key. Service charges may apply under your account terms.
  • Check that the selected input captures the required voices and the service is reachable. This route streams audio and does not require a local recognition model download.

First use: configure in order

  1. Open Setup → API configuration. Choose Doubao Speech · Realtime recognition. This is a separate speech connection, not an Ark text connection. Existing Ark settings do not automatically supply its credentials.
  2. Save the service address and speech API key. Use the HTTPS root https://openspeech.bytedance.com and the API key issued by the current speech console. Do not substitute an Ark endpoint or text-service key, or append an OpenAI-compatible path.
  3. Select an enabled ASR resource. volc.seedasr.sauc.duration or volc.seedasr.sauc.concurrent must match your account access. Configure the resource here once, then finish the wizard and save.
  4. Choose the saved service in Recording settings. Set microphone recognition to Realtime API and choose Doubao / Doubao Speech as the saved service. The saved address, resource and key are reused. Once its ASR resource is saved in Setup, it is shown read-only here; do not enter or change it again on the recording page.
  5. Set source language to Automatic and leave language hints empty. This bidirectional API does not support forcing a recognition language. Change a previous fixed source language such as Chinese or English back to Automatic. A translation target belongs to a separate configuration.
  6. Speak a sentence, then Stop and wait. Return to live subtitles and click Start. Interim text appears while you speak; the service then confirms final sentences. After stopping capture, wait for the remaining confirmed results and task-completion indication.
Actual production UI with controlled IPC configuration, copied without image edits. This does not establish live-cloud, billed-account, recognition-quality or physical-recording acceptance.

Change a resource in Setup

  1. In the Realtime API area of Recording settings, click Manage service. It opens Setup directly at API configuration.
  2. Find Doubao Speech · Realtime recognition, update its ASR resource, address or speech API key, then finish the wizard and save. Use the enabled resource volc.seedasr.sauc.duration or volc.seedasr.sauc.concurrent that matches your account.
  3. Return to Recording settings, confirm Doubao Speech is selected, and check the displayed resource. Saved changes apply to the next recording. An active recording does not switch services or resources midway.

If “The saved model has changed” appears, refresh the settings or reselect the service before starting a new recording. If the displayed resource is invalid or unavailable to your account, correct it through Manage service and save. The app does not silently substitute another resource or resend audio.

How this route works

Server segmentationSegmentation is handled by Doubao; local Silero or manual sentence splitting does not drive this route.
Interim text changesUse confirmed sentences for review. Drafts are not promoted to final text.
Completion after StopStopping microphone capture still requires final server confirmation.

A confirmed final sentence received after Stop still reaches captions and the meeting record. It is also submitted for translation when translation is enabled. No further text is typed into other windows after Stop. Wait for this task to finish before starting another recording; unconfirmed drafts are not added as final text.

If something goes wrong

Why configure speech separately from Ark?
The services use different addresses, credentials and resources. This route uses the dedicated bigmodel_async bidirectional audio protocol, not an OpenAI-compatible text API. Open API configuration through Manage service and save the speech connection there; recording pages only select it.
The resource is read-only or unsupported
This is the resource saved in Setup. Choose Manage service, set ASR to the full enabled duration or concurrent resource ID, save, then return and reselect the service. Do not use an Ark model name.
A language-setting error appears
Set the recording source language to Automatic, clear language hints, then start a new recording. This API cannot force recognition by specifying zh or en.
Access fails or no confirmed result arrives
Check the speech-console key, enabled resource, account status and connection. Changing Ark text models does not resolve this speech configuration.
The connection fails or completion is interrupted
The original recording and confirmed results are retained; unconfirmed drafts do not become final text. The app does not automatically retry, reconnect or resend audio. After addressing the error, manually import the saved recording into the separate file-transcription workflow if needed, checking its service and possible charges.

Scope

This integration adds Doubao realtime ASR/STT only. It does not establish full speech-to-speech interpreting, cloning, synthesis or a file-upload API. Translation, translated audio and Linglu reports retain their own services and settings.

Official references: Bidirectional realtime recognition · Streaming recognition API