Drop the Bass with Embedding and Vectors in Azure AI Search
summary
Demos show embedding pipelines that actually run: ResNet18 with PyTorch for image features, VGGish embeddings for audio, Annoy and Azure AI Search for nearest‑neighbor queries. A 12,300‑item dog index and 159,000 music phrases get vectorized; image and face similarity searches work well, face demo later restricted by the EU AI Act, pure “similar‑song” embeddings fail to mix, but combining chroma features, BPM normalization to 126 and rubberband key shifting produces usable Azure AI Search vector index results for harmonic mixing.
AI-written summary · about
chapters (10)
- 0:00Welcome
- 2:50Why embeddings and vectors
- 6:42Text embedding and cosine similarity
- 12:18Image embedding with neural networks
- 18:41Face recognition demo with vector search
- 22:11Music mixing with vector search
- 31:55Music theory and harmonic mixing
- 41:39Extracting phrases and embeddings for music
- 56:08Challenges and lessons learned
- 1:01:20Live coding demo and Q&A
related talks
01h 03mDrop the Bass with Embedding and Vectors in Azure AI Search
46mAdvanced RAG with Azure AI Search - top questions answered
15mAzure Cosmos DB as Vector Database
50mBuilding AI ready applications
44mSupercharge your data with Azure AI Search and OpenAI
12mThe mechanics of integrating AI and vector search into Azure Database for PostgreSQL
44mVector Search Isn’t Enough
11mSimilarity Search Application utilizing Vector types on vCore-based Azure Cosmos DB for MongoDB