title	Timbre Text to Emotion
emoji	🎭
colorFrom	purple
colorTo	pink
sdk	gradio
sdk_version	5.25.0
python_version	3.10
app_file	app.py
pinned	false

Timbre Text-to-Emotion

Classifies a free-text mood description into one of Timbre's 16 emotion categories and returns a target acoustic profile. Used by the Timbre brief engine when a client types a mood description during onboarding instead of uploading a reference track.

Space: WCA0202/Timbre-Text-to-Emotion GitHub: willwang0202/Timbre-Text-to-Emotion

---

API

Single endpoint exposed via the Gradio SSE pattern:

POST /call/classify_mood
Body: { "data": ["your mood description here"] }
→ { "event_id": "abc123" }

GET /call/classify_mood/{event_id}
→ SSE stream — read "event: complete" then the next "data:" line

Response

{
  "emotion": "melancholic",
  "confidence": 0.87,
  "acoustic_profile": {
    "valence": 3.5,
    "arousal": 3.2,
    "bpm": 72,
    "mood_happy": 0.08,
    "mood_sad": 0.72,
    "mood_aggressive": 0.05,
    "mood_relaxed": 0.55,
    "mood_party": 0.03,
    "danceability": 0.20
  }
}

On error: { "error": "<message>" }

Emotion taxonomy (16 categories)

euphoric, uplifting, passionate, melancholic, bittersweet, longing, aggressive, tense, rebellious, anxious, mysterious, dark, calm, nostalgic, dreamy, romantic

---

Model

j-hartmann/emotion-english-distilroberta-base — DistilRoBERTa fine-tuned on the MELD, EmotionLines, ISEAR, WASSA, CrowdFlower, and GoEmotions datasets.

• Size: ~82 MB
• Labels: anger, disgust, fear, joy, neutral, sadness, surprise
• Runtime: CPU-only (HF free tier)

The 7 model labels are mapped to Timbre's 16 emotions via a weighted blend table in emotion_classifier.py. Multi-label inputs produce meaningful compound outputs (e.g. a text scoring high on both joy and sadness maps toward bittersweet).

---

Files

File	Purpose
app.py	Gradio interface — classify_mood endpoint, model warm-up on startup
emotion_classifier.py	Loads model, weighted label→emotion mapping, baked-in EMOTION_PROFILES
requirements.txt	gradio==5.25.0, transformers==4.40.2, torch==2.2.2, numpy

---

Running locally

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python app.py   # http://localhost:7860

The model (~82 MB) is downloaded from HF Hub on first run and cached in ~/.cache/huggingface/.

---

Deploying

Push to both remotes to keep GitHub and HF Space in sync:

git push origin main   # → GitHub
git push hf main       # → HF Space