Spaces:

---

Duplicated from  facebook/MusicGen

sylus256

/

MusicGen

like 0

Build error

Build Luganda singing pipeline for BeatDrop Studio using Makerere/Sunbird + RVC.

1. Create API route /api/luganda-sing:

   • Input: { lyrics: string, style: "kidandali"|"afrobeats"|"ballad", use_my_voice: boolean }

   • Step A - Get perfect Luganda pronunciation:
     POST https://api-inference.huggingface.co/models/Sunbird/sunbird-lug-tts
     Headers: { Authorization: "Bearer " + process.env.HF_TOKEN, "Content-Type": "application/json" }
     Body: { inputs: lyrics }
     Response is audio/wav - save as /tmp/base.wav

   • Step B - If use_my_voice = true AND model sylus256/luganda-beatdrop exists:
     POST https://api-inference.huggingface.co/models/sylus256/luganda-beatdrop
     Headers same as above
     Send base.wav as file with parameters: { f0_up_key: 0, index_rate: 0.75, filter_radius: 3 }
     This converts Sunbird speech to your singing voice
     Else: use base.wav directly

   • Step C - Return JSON: { audio_url: data:audio/wav;base64,..., source: "sunbird+rvc" }

2. Update /studio page:

   • Add textarea "Lyrics in Luganda" with placeholder "Nkwagala nyo BeatDrop..."
   • Add toggle "Sing with my voice" (default ON)
   • Add dropdown style: Kidandali, Afrobeats, Ballad
   • Button "Generate Luganda Singing" calls /api/luganda-sing
   • Show audio player + download button
   • Show credit: "Powered by Makerere Sunbird AI + BeatDrop Voice"

3. Cache results in Supabase table 'generations':

   • columns: id, user_id, lyrics, style, audio_url, created_at
   • If same lyrics+style exists in last 24h, return cached

4. Add fallback: If Sunbird model is loading (503), retry 3 times with 5s delay, show "Makerere model warming up..."

IMPORTANT: Log each step to console for debugging.