LobeHub
LobeHub

Using Doubao Speech in LobeHub

Doubao Speech is ByteDance Volcengine's speech synthesis service. Its Seed-TTS 2.0 model speaks natural Mandarin and English with official voices. In LobeHub, Doubao Speech is a text-to-speech provider only: it has no chat models.

Doubao Speech is a separate service from Volcengine Ark (the Volcengine provider for Doubao chat, image and video models), with its own console and credentials.

Step 1: Obtain your Doubao Speech credentials

  • Sign in to the Volcengine speech console and enable Doubao Speech Synthesis 2.0.
  • Recommended: in the new console's API Key management, create an API key and copy it.
  • Legacy console accounts: copy the App ID and Access Token of your application instead.

Step 2: Configure Doubao Speech in LobeHub

  • Open App Settings → AI Providers and select Doubao Speech (type Doubao in the provider search box to find it).
  • Paste the API key into API Key, or fill in App ID (legacy console) and Access Token (legacy console). When both are filled in, the API key is used. Changes are saved automatically, and the credentials are encrypted with AES-GCM before they are stored.
  • Turn on the switch next to the provider name.
  • Optional: to send requests through a proxy, enter it in API Proxy URL. Leave it empty to use https://openspeech.bytedance.com.

Step 3: Choose models

The Model List shows Doubao Speech 2.0 (seed-tts-2.0) under the TTS tab, enabled by default.

Doubao Speech has no chat model, so the provider page shows no Check button. Verify the credentials by generating speech as described below.

Doubao Speech Synthesis 2.0 is billed at 3 CNY per 10,000 characters; every character, punctuation mark and space counts as one. Review the Volcengine speech pricing before generating large amounts of audio.

Generate speech

Speech synthesis runs through the LobeHub text-to-speech API, which the LobeHub CLI (lh) and the OpenAI-compatible POST /api/v1/audio/speech endpoint both use:

*.sh
Shell
lh generate tts "你好,LobeHub" -m doubaospeech/seed-tts-2.0 -o hello.mp3
  • Voice — when no voice is given, Vivi 2.0 (zh_female_vv_uranus_bigtts) is used. Pass --voice <voice_type> to use another official 2.0 voice, for example zh_male_m191_uranus_bigtts (云舟) or en_male_tim_uranus_bigtts (Tim), or a voice clone of your own (S_…), which is sent to the voice-cloning resource seed-icl-2.0 automatically and needs Doubao Voice Cloning 2.0 enabled. 1.0 voices (*_moon_bigtts, *_mars_bigtts) do not work with the 2.0 model.
  • Length — each request is limited to 4,096 characters, because the audio is returned in a single response. Split longer text into several requests.
  • Format — mp3 (default) or opus (always 48 kHz).
  • Speed and volume — 0.5 to 2.
  • Pitch — -12 to 12 semitones.
  • Sample rate — 8000, 16000, 22050, 24000 (default), 32000, 44100 or 48000 Hz.

Troubleshooting

"Doubao Speech API Key is incorrect or empty, please check your Doubao Speech API Key and try again." — No credentials were found, only one of App ID and Access Token was filled in, or Volcengine rejected them. Enter a valid API key (or App ID + Access Token) in App Settings → AI Providers → Doubao Speech, or ask the administrator to set DOUBAOSPEECH_API_KEY on the server.

"Enable the voice for this account…" — The voice is not enabled for your account. Enable it in the Volcengine speech console, or pick another voice.

"The voice does not belong to this model version…" — A 1.0 voice was used with the 2.0 model, or Doubao Speech Synthesis 2.0 is not enabled for the account.

Rate-limit or quota errors — The account reached its concurrency limit (10 by default) or ran out of balance or trial quota. Check the Volcengine console.

For self-hosted deployments, see DOUBAOSPEECH_API_KEY in the model provider environment variables.