Clone yourself with Azure Custom Neural Voice
summary
Azure Custom Neural Voice lets you fine tune Microsoft’s neural TTS with either 20–50 utterances (light, ~1.5 hours total to record + train) or a pro workflow with 300–2,000+ utterances and ~24+ hours training; pro requires approval and talent consent. Training costs run about $52 per compute hour and models incur ~$4/hour when deployed. Speech Studio handles recording, lexicons/IPA fixes, and bulk uploads (podcast MP3s), and light models often sound usable even from a laptop mic.
AI-written summary · about
chapters (26)
- 0:00Intro
- 1:37Clone yourself
- 5:12Open AI
- 7:16Human Noise Box
- 11:26Computer Talk
- 16:00AI ML Models
- 19:55Azure Custom Voice
- 22:44Responsible Use
- 24:51Training
- 26:10Recommendations
- 26:54Azure Speech Studio
- 28:44Naming your model
- 31:19Demo
- 32:39monotone demo
- 33:09congratulations
- 34:19accents
- 36:19French model
- 36:49Deployed model
- 38:19Light vs Pro
- 40:39Custom Neural Voice
- 41:11Training Data
- 41:59Accuracy
- 42:59Speech Studio
- 48:29Real use cases
- 50:49Cost
- 53:46Wrap up
related talks
01h 06mClone yourself with Azure Custom Neural Voice
58mClone yourself with Azure Custom Neural Voice
46mEnhance your solutions with new Azure AI products and features
01h 29mAzure AI Foundry Survival Guide
49mMake your applications intelligent (...)
51mCan AI Face Your (Potential) Customers?
01h 10mSolutions prototyping using Azure AI Studio
17mServing Voice AI at Scale