SenseAudio Polyphone TTS (多音字)

Precise pronunciation control for Chinese TTS via pinyin annotation. The dictionary parameter lets you override how specific characters are read — essential for polyphones (多音字) that the model might guess wrong.

The dictionary parameter only works with cloned voices and model SenseAudio-TTS-1.5. System voices (male_0004_a etc.) do not support it.

Step 1: Scan for Polyphones

When the user provides text, scan it for these common polyphones and flag any that appear:

Show the user which polyphones were found and your best guess at the intended reading, then ask them to confirm or correct before synthesizing.

Example:

检测到多音字：
- "行" (第2个): 银行 → 建议读 háng [hang2] ✓ 还是 xíng [xing2]?
- "行" (第4个): 行长 → 建议读 zhǎng [zhang3] ✓ 还是 cháng [chang2]?

Step 2: Build the Dictionary

Convert confirmed readings into the dictionary array. Each entry covers one phrase containing the polyphone:

原文片段 → replacement 格式：在多音字前加 [pinyin]，其余字保持原样

Pinyin format: [声母韵母声调数字] — e.g., [hang2]、[xing2]、[zhang3]

Example:

• original: 银行行长
• replacement: 银[hang2]行[zhang3]长

Build the full dictionary array:

"dictionary": [
  {"original": "银行行长", "replacement": "银[hang2]行[zhang3]长"},
  {"original": "好奇心", "replacement": "[hao4]奇心"}
]

Each original should be a short phrase (3–8 chars) that uniquely identifies the occurrence in context. Avoid single-character originals — they may match unintended occurrences.

Step 3: Synthesize

The user must provide a cloned voice ID. If they don't have one, remind them that dictionary requires a cloned voice and suggest using the senseaudio-voice-cloner skill first.

curl -s -X POST https://api.senseaudio.cn/v1/t2a_v2 \
  -H "Authorization: Bearer $SENSEAUDIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "SenseAudio-TTS-1.5",
    "text": "<TEXT>",
    "stream": false,
    "voice_setting": {
      "voice_id": "<CLONED_VOICE_ID>"
    },
    "audio_setting": {
      "format": "mp3"
    },
    "dictionary": <DICTIONARY_ARRAY>
  }' -o response.json

jq -r '.data.audio' response.json | xxd -r -p > output.mp3

Check base_resp.status_code == 0 before decoding.

Step 4: Iterate

After the user listens, they may find additional mispronunciations. Update the dictionary array and re-synthesize. Keep the previous response.json until the new one succeeds.

Report: file path, duration (jq '.extra_info.audio_length' response.json ms), character count, and which dictionary entries were applied.