SenseAudio Lyric Flip

Rewrite a song with a new theme while keeping the original's structure. The key insight: AI-generated lyrics drift without constraints. By extracting a structural skeleton first and using it as a hard constraint, the output stays tight and singable.

Step 1: Collect Inputs

Ask the user for:

1. 参考歌词 — paste the original lyrics (or a verse/chorus excerpt)
2. 新主题 — what the new song should be about (e.g., "程序员加班", "失恋的猫", "创业失败")
3. 风格保留还是改变？ — keep the original vibe, or shift it (e.g., "结构照搬但从抒情改成摇滚")

Step 2: Extract the Skeleton (LLM task)

Analyze the reference lyrics yourself — no API call needed for this step. Extract:

Structure map:

• Segment labels: verse / chorus / bridge / pre-chorus / outro
• Number of lines per segment
• Character count range per line (e.g., 7–9 chars)

Rhyme scheme:

• Label each line's end sound: A, B, C...
• Example: AABB means lines 1&2 rhyme, lines 3&4 rhyme
• Note which segments rhyme internally vs. across segments

Rhythm feel:

• Syllable density: sparse (≤6 chars/line) / medium (7–10) / dense (11+)
• Repetition patterns: does the chorus repeat a phrase? Does a hook line recur?
• Emotional arc: builds up / stays level / drops at bridge

Example skeleton output to show the user:

段落结构：
  [verse] × 2 (各4行，7-9字/行，ABAB韵)
  [chorus] × 2 (各4行，6-8字/行，AABB韵，第4行重复)
  [bridge] × 1 (2行，自由韵)

情绪走向：verse 平静叙述 → chorus 情绪爆发 → bridge 转折收尾

韵脚示例（verse）：
  行1: __韵A
  行2: __韵B
  行3: __韵A
  行4: __韵B

Show this to the user and confirm before proceeding.

Step 3: Generate New Lyrics via API

Build a tightly constrained prompt using the extracted skeleton:

请根据以下结构约束，为主题"\x3C新主题>"创作歌词：

段落结构：\x3C从骨架提取的结构>
每行字数：\x3C范围>
韵脚模式：\x3CAABB/ABAB等>
情绪走向：\x3C从骨架提取>
风格：\x3C保留原风格 或 用户指定的新风格>

硬性要求：
- 严格遵守每段行数
- 每行字数在指定范围内
- 韵脚必须对齐（同一韵组的行必须押韵）
- 不要超出段落结构，不要添加额外段落

LYRICS_RESP=$(curl -s -X POST "https://api.senseaudio.cn/v1/song/lyrics/create" \
  -H "Authorization: Bearer $SENSEAUDIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"prompt\": \"\x3CPROMPT>\", \"provider\": \"sensesong\"}")

TASK_ID=$(echo $LYRICS_RESP | jq -r '.task_id // empty')

Poll if async:

while true; do
  POLL=$(curl -s "https://api.senseaudio.cn/v1/song/lyrics/pending/$TASK_ID" \
    -H "Authorization: Bearer $SENSEAUDIO_API_KEY")
  STATUS=$(echo $POLL | jq -r '.status')
  [ "$STATUS" = "SUCCESS" ] || [ "$STATUS" = "FAILED" ] && break
  sleep 3
done
LYRICS=$(echo $POLL | jq -r '.response.data[0].text')

Step 4: Rhyme Check

After getting the lyrics, verify rhyme alignment yourself before showing the user. For each rhyme group in the skeleton:

• Extract the last 1–2 characters of each line in the group
• Check if they share a vowel sound (approximate check is fine)
• Flag lines that clearly don't rhyme with their group

Show the user a marked-up version:

[verse]
行1: 深夜还在敲代码 ✓ (韵A: -ā)
行2: 需求改了又改了 ✓ (韵B: -le)
行3: 眼睛快要睁不开 ✓ (韵A: -āi ≈ ā)
行4: 产品说明天上线了 ⚠ (韵B: -le ✓ 但字数偏长)

If rhyme issues are minor, note them and ask if the user wants to regenerate those lines. If structure is badly off, regenerate the full lyrics with a stricter prompt.

Step 5: Compose the Song

Once lyrics are approved, infer the style from the reference song and user preferences:

If the user wants to shift the style, use their specified direction instead.

SONG_RESP=$(curl -s -X POST "https://api.senseaudio.cn/v1/song/music/create" \
  -H "Authorization: Bearer $SENSEAUDIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{
    \"model\": \"sensesong\",
    \"lyrics\": \"\x3CAPPROVED_LYRICS>\",
    \"title\": \"\x3C新主题 + 仿 原曲名>\",
    \"vocal_gender\": \"\x3Cf|m>\",
    \"style\": \"\x3CINFERRED_STYLE>\"
  }")

SONG_TASK=$(echo $SONG_RESP | jq -r '.task_id')

Poll until done (30–120s):

while true; do
  POLL=$(curl -s "https://api.senseaudio.cn/v1/song/music/pending/$SONG_TASK" \
    -H "Authorization: Bearer $SENSEAUDIO_API_KEY")
  STATUS=$(echo $POLL | jq -r '.status')
  [ "$STATUS" = "SUCCESS" ] || [ "$STATUS" = "FAILED" ] && break
  echo "作曲中..."
  sleep 5
done

Output

改编完成：

原曲结构：\x3C骨架摘要>
新主题：\x3C用户输入>
风格：\x3C使用的曲风>

歌词预览：
\x3C最终歌词>

音频：\x3Caudio_url>
封面：\x3Ccover_url>
时长：\x3Cduration> 秒

If the user wants to iterate — adjust rhymes, change a line, shift the style — update the lyrics and re-submit to the music API. The skeleton stays fixed; only the words change.