Doubao TTS Skill (V3)

Use the Volcengine Doubao Text-to-Speech V3 HTTP SSE unidirectional streaming API to convert text into audio files.

Get an API Key (Recommended, New Console)

1. Log in to the Volcengine Console
2. Go to Doubao Voice → Text-to-Speech Large Model → Application Management
3. Create an application or use an existing one
4. On the API Key Management page, get the API Key → corresponding to DOUBAO_TTS_API_KEY

If you have not enabled the service yet, enable it first on the Text-to-Speech Large Model page.

Legacy Console (AppID + Token)

At the bottom of the application details page in the legacy console, you can find:

• APP ID → DOUBAO_TTS_APPID
• Access Token → DOUBAO_TTS_TOKEN

Environment Variables

Check whether they are configured:

[ -n "$DOUBAO_TTS_API_KEY" ] && echo "API_KEY: set" || echo "API_KEY: not set"
[ -n "$DOUBAO_TTS_APPID" ] && echo "APPID: set" || echo "APPID: not set"
[ -n "$DOUBAO_TTS_TOKEN" ] && echo "TOKEN: set" || echo "TOKEN: not set"

If they are not configured, tell the user to set them (prefer API Key): Set DOUBAO_TTS_API_KEY | Set DOUBAO_TTS_RESOURCE_ID

Legacy console (AppID + Token): Set DOUBAO_TTS_APPID | Set DOUBAO_TTS_TOKEN

Usage

Call the script: /var/minis/skills/doubao-tts/scripts/tts.py

# Basic usage
uv run --script --cache-dir /root/.cache/uv \
  /var/minis/skills/doubao-tts/scripts/tts.py \
  --text "Hello, and welcome to Doubao Text-to-Speech." \
  --output /var/minis/workspace/output.mp3

# Specify voice and speech rate
uv run --script --cache-dir /root/.cache/uv \
  /var/minis/skills/doubao-tts/scripts/tts.py \
  --text "The weather is really nice today." \
  --speaker zh_female_cancan_uranus_bigtts \
  --speech-rate 10 \
  --output /var/minis/workspace/output.mp3

# English
uv run --script --cache-dir /root/.cache/uv \
  /var/minis/skills/doubao-tts/scripts/tts.py \
  --text "Hello! Nice to meet you." \
  --speaker en_female_dacey_uranus_bigtts \
  --output /var/minis/workspace/output.mp3

API Description

• Endpoint: https://openspeech.bytedance.com/api/v3/tts/unidirectional/sse (SSE streaming)
• Authentication (choose one):
  • New console: Header X-Api-Key (API Key)
  • Legacy console: Header X-Api-App-Id + X-Api-Access-Key (AppID + Token)
• Resource ID: Specifies the model version to call (see the table below)
• Usage response: The script includes X-Control-Require-Usage-Tokens-Return: text_words by default, which returns the number of billable characters (text_words) when synthesis ends

Parameters

Quick Reference for Common Voices

Doubao Text-to-Speech Model 2.0 (seed-tts-2.0, recommended)

Doubao Text-to-Speech Model 1.0 (seed-tts-1.0, requires changing --resource-id)

⚠️ 1.0 and 2.0 voices cannot be mixed. seed-tts-2.0 supports only voices ending in *_uranus_bigtts.

Common Emotion Values

pleased (pleased) / sorry (sorry) / happy (happy) / sad (sad) / angry (angry) / scare (scared) / surprise (surprised) / hate (disgust) / tear (tearful voice) / narrator (narrator) / storytelling (storytelling)

Complete Workflow

1. Check whether environment variables are configured (first DOUBAO_TTS_API_KEY, then DOUBAO_TTS_APPID + DOUBAO_TTS_TOKEN)
2. Call the tts.py script to generate an audio file in /var/minis/workspace/
3. Return it to the user as a minis://workspace/xxx.mp3 link, which they can click to play directly.